HCSA-Presales-Storage V4.0 Training Material

Basic Knowledge of Data Storage
0 Huawei Confidential
Contents
1. Positioning and Classification of Data Storage
2. Basic Concepts of Disks
3. Basic Concepts of Disk Redundancy Protection
4. Data Storage Development Trend
New Data Center: New Compute, Storage, Network, and Energy
Layer 4 Application
• ERP • iCare • W3
APPLICATION • PDM • DW • Email
• iSales • HRMS • ...
Service platform
Layer 3 Database, big data, and middleware
Application middleware IT • WAS • Message middleware • LDAP
Database and big data arc • Oracle DB • MySQL • DNS
hit • Integration via EIP • SQL Server • Big data
Compute ect
Layer 2 Compute, storage, and network
Storage
ure
lay • Compute • Storage • Network
Network ers
Layer 1 Energy
Equipment room • Electric power • Cabling system • Physical
system • Fire extinguishing security
• Cooling system system • ...
Reliable Storage Systems of High Performance and Large Capacity Dedicated
to Data Centers
Storage Products for Enterprise Data
Storage Products for Individuals
Centers
• Requirements: cost-effectiveness and fair and • Requirements: data integrity, high reliability, high
reasonable prices performance, and large capacity
• Different from servers in high availability (HA)
• Tolerance for disadvantages in response time, service
architecture, power failure protection, and quality
life, and errors
control
Storage System Classification by Service – Block Storage
iSCSI or Fibre Channel
Protocol layer ⚫ Block storage is used to store structured data, that is,
data is directly read and written by reading or writing
one or more addresses from or into storage space.
⚫ Generally, block storage is used to store important data
Storage of virtualization applications such as Oracle database
layer ...
and VMware.
Block storage:
• Advantages: direct access, minimized overhead, and
highest efficiency
• Disadvantages: highest cost and poor scalability
• Application scenarios: Oracle database and VMware
virtualization
Storage System Classification by Service – File Storage
NFS or CIFS
⚫ File storage is mainly used to store unstructured data.

File system Professional file systems are added to block storage devices
to implement file sharing.
⚫ Generally, file storage is mainly used to store file data, such
as PACS (medical imaging) data, carrier CDR data,
manufacturing (EDA) simulation data, oil exploration data,
... and HPC computing data.
File storage:
• Advantages: easy management and interconnection with applications
• Disadvantage: supports expansion, requiring ecosystem compatibility.
• Application scenarios: enterprises' internal application integration and file
sharing
Storage System Classification by Service – Object Storage
HTTP, REST, SOAP, or S3
⚫ Object storage, also known as object-based storage

Object (OBS), is a network storage architecture. The differences
Object Object Object between OBS and block storage or file storage lie in the
Object
storage Key Data interfaces (S3 interfaces) provided by OBS. OBS only
Object
Object generates an ID for the metadata of stored data and
Metadata
stores the ID, regardless of the data type.
Customized
⚫ This storage architecture is mainly used in the
metadata application scenarios that have low requirements on
...
performance but high requirements on capacity. It is
mainly used to meet customers' requirements on large
capacity and low price. The application scenarios include
Features of object storage: public cloud, Internet, and space leasing.
• Flat structure and nearly unlimited capacity expansion
• More intelligent self-management
• Standard Internet protocols and cross-region transmission
capabilities
• Application scenarios: Internet-oriented storage, archiving, and
backup
Storage System Classification by Architecture – Centralized Storage
Controller enclosure Controller enclosure

Controller A Controller B Controller A Controller B Centralized storage:
Backplane Backplane • Controller: manages all data. When a
interconnection interconnection controller is faulty, another controller takes
Cache Cache Cache Cache over services.
Horizontal • Disk enclosure: only stores data. The
scaling
controller determines how to store data.
Disk enclosure 1 Disk enclosure 1 • The capacity generally ranges from TB to
dozens of PB.
• Controllers in a controller enclosure are
...
...
Scale up directly connected through backplane circuits,

Disk enclosure N Disk enclosure N and controller enclosures are interconnected
through networks.
Note: Devices that support multiple controllers can be expanded horizontally.
Generally, a maximum of 32 controllers are supported.
Storage System Classification by Architecture – Scale-Out Storage
Horizontal scaling: 3 nodes at least,

scalable to thousands of nodes
Node 1 Node 2 Node 3 Node N

Scale-out storage:
Cache Cache Cache Cache • Each node participates in data management and storage.
If one node is faulty, the remaining nodes continue to
... work.
Disk Disk Disk Disk • The capacity ranges from 100 TB to EB.
• Nodes are interconnected through the network.
Switch 1 Switch 2
Contents
Disk Types
Size Interface protocols

3.5-inch
ATA, IDE
2.5-inch
SATA, NL-SAS, SAS
1.8-inch
Fibre Channel, NVMe
Function
Media
HDD
Disk types Desktop level
Monitoring level
SSD Enterprise level
Enterprise-level disks are designed for enterprise applications, such as storage arrays and
servers. Desktop-level disks are mainly used for home applications, such as desktop PCs and
laptops.
Classification by Media
⚫ A hard disk drive (HDD) is a traditional hard disk. It consists of a platter,

a head, a platter rotating shaft, a control motor, a head controller, a
data converter, interfaces, and cache.
⚫ A solid-state disk or solid-state drive (SSD) is also called an electronic

disk or a solid-state electronic disk. Different from HDDs that use
mechanical parts such as disk bodies and heads, SSDs are composed of
control chips and storage chips (flash or DRAM chips). In other words,
SSDs are made of solid-state electronic storage chip arrays.
SSDs vs. HDDs
Item SSD HDD

Storage media Flash chip Magnetic disk
Shockproof and drop resistance High Low
Data storage speed High: xxx, xxx IOPS, hundreds of MB/s Low: 200 to 300 IOPS, dozens of MB/s
Power consumption Low High
Weight Light Heavy
Noise None Yes
Capacity xxx GB to 32 TB x TB
Service life Long Relatively long
By Protocol – Serial ATA (SATA)
Advantages:
1. SATA is a serial bus interface protocol. During data transmission, the data and signal cables are used independently,
and the embedded clock frequency signal is used. Its rate can reach 30 times of the Parallel ATA (PATA).
2. Instead of simple PATA improvement, the bus structure is brand new.
3. Control information is scattered in data and transmitted by using a predefined bit.
4. One path is used to transmit data, and the other is used to return a response.
5. SATA has higher anti-interference capabilities and faster speed than PATA. The installation is easier and the number of
cables used in the chassis is reduced.
6. The performance of single-thread tasks is good.
Disadvantages:
1. SATA, designed for entry-level applications, is not as powerful as SCSI in terms of big data throughput or multi-thread
transmission.
2. When multiple threads are reading data, the head of a disk swings back and forth, causing the disk to be overheated.
By Protocol – Serial Attached SCSI (SAS)
1. SCSI is a storage interface protocol specially designed for small computer systems. 50-pin ports are used. The
appearance of a 50-pin port is similar to that of a common disk port. The SCSI disks support connections to various
devices. In addition, each SCSI disk has an independent chip for data processing. The CPU usage is low, the
bandwidth can reach 320 MB/s, and the stability is good.
2. SAS is a disk connection technology that integrates the advantages of the parallel SCSI and serial connection
technologies.
3. SAS is a point-to-point, full-duplex, and dual-port interface.
4. SAS is compatible with SATA, delivering high performance for enterprises, achieving interoperability with SATA, and
bringing unprecedented flexibility and benefits to enterprises.
5. SAS features high performance, high reliability, and powerful scalability.
The serial technology is also used, which is better than SCSI in transmission rate and anti-interference. However, the
price is higher.
By Protocol – Near-Line SAS (NL-SAS)
1. NL-SAS disks integrate SAS interfaces and SATA disks. In other words, NL-SAS disks refer to SATA disks
with SAS interfaces and near-SAS performance.
2. Near-line storage is oriented to applications between online storage and offline storage. Data that is not
frequently used or seldom accessed is stored on the storage devices of which performance is relatively
low. However, these devices must provide fast addressing capabilities and a high transmission rate.
SAS SSD
Size: standard 2.5-inch SAS SSDs accessed based on the SAS protocol
Short access time: The xx μs-level average read/write response time meets the
requirements of latency-sensitive service applications and minimizes the response time
to customers' requirements.
High throughput: Provides hundreds of thousands of random read/write IOPS, meeting
the requirements of high-performance storage system.
High read/write speed: GB-level read/write bandwidth and stable performance.
Low power consumption: The energy consumption per unit capacity is much lower than
that of SAS HDDs.
NVMe SSD
Size: 2.5-inch and 1.8-inch NVMe SSDs accessed based on the NVMe
protocol
Short access time: The xx μs-level average read/write response time,
shorter than that of SAS SSDs, meets the requirements of latency-
sensitive service applications and minimizes the response time to
customers' requirements.
High throughput: Provides hundreds of thousands of random read/write
IOPS, meeting the requirements of high-performance storage system.
High read/write speed: GB-level read/write bandwidth and stable
performance.
Low power consumption: The energy consumption per unit capacity is
much lower than that of SAS HDDs.
Comparison of Mainstream Disk Types
Indicator SATA HDD SAS HDD NL-SAS HDD SAS SSD NVMe SSD
Rotational speed
7200 15,000 or 10,000 7200 N/A N/A
(rpm)
Capacity (TB) 4, 6, 8, 10, or 14 1.2, 1.8, or 2.4 4, 6, 8, 10, or 14 1.92, 3.84, 7.68, 15.36, etc. 1.92, 3.84, 7.68, 15.36, etc.
MTBF (h) 1,200,000 1,600,000 1,200,000 2,000,000 2,000,000
An SSD is made up of a solid-state

An SSD is made up of a solid-state
Being developed from SAS disks are designed to meet NL-SAS disks are electronic storage chip array. An SSD
electronic storage chip array. An
ATA disks, SATA 2.0 enterprises' high performance enterprise-class SATA consists of a control unit and a storage
SSD consists of a control unit and
supports 300 MB/s requirements and are drives with SAS unit (flash and DRAM chips). SSD is the
a storage unit (flash and DRAM
data transfer, and compatible with SATA disks. interfaces. They are same as the common disks in the
chips). SSD is the same as the
SATA 3.0 supports up The transmission rate ranges applicable to tiered regulations and definition of interfaces,
Remarks common disks in the regulations
to 600 MB/s data from 3.0 Gbit/s to 6.0 Gbit/s, storage in a disk array, functions, and usage.
and definition of interfaces,
transfer. and will be increased to 12.0 which simplifies the The latency is lower than that of SAS SSDs,
functions, usage, as well as the
The annual failure rate Gbit/s in the future. design of the disk array. and the size can be thinner than that of
exterior and size.
of SATA disks is about The annual failure rate of SAS The annual failure rate of standard SAS SSDs.
The annual failure rate is about
2%. disks is less than 2%. NL-SAS disks is about 2%. The annual failure rate is about 0.5% to
0.5% to 0.8%.
0.8%.
SCM – Next-Generation Storage Media
Storage Class Memory (SCM) is a new storage medium popular in the High-performance SSDs provide the following application forms
industry. SCM is similar to storage in its persistence and to memory in in a storage system:
its byte-level access. • Metadata cache: As metadata cache of AFA, SCM SSDs work
with the DRAM to build a memory + SCM SSD two-tier
The SCM SSD, which uses NVMe block interface and is compatible cache, which avoids bottlenecks in memory capacity and
with the native architecture, is the primary application form of SCM. supports larger user capacity with stable performance.
Optane P4800X series launched by Intel is an example. This product • Data cache: SCM SSDs serve as the acceleration layer of user
has little impact on the system architecture but provides better data and improves performance in typical application
performance than that of flash SSDs. In addition, SCM SSDs do not scenarios.
require garbage collection, which prevents performance deterioration • Main storage: SCM SSDs serve as the storage layer of user
similar to that of NAND SSD after long-time running, and curbs data and provides a high-performance storage system to
latency at an appropriate level. This type of storage products is meet the performance requirements of some scenarios.
springing up in the industry, and Huawei is also working on the R&D
of relevant products.
Limited by the price of SCM, its high requirements on the storage architecture, and unclear application scenarios, it
is difficult to put SCM into large-scale commercial use in a short term.
Contents
RAID
The term RAID was invented by David Patterson, Garth A. Gibson, and Randy Katz at the University of California, Berkeley
in 1987. It combines multiple independent physical disks into a virtual logical disk using related algorithms to provide
larger capacity, higher performance, and better error tolerance capabilities.
RAID10 RAID50
⚫ RAID is classified into different RAID levels based
on the combination methods, such as RAID0,
RAID1, RAID3, RAID5, and RAID6.
⚫ Two different RAID levels can be combined to form
a new RAID level.
Comparison between Mainstream RAID Levels
Level RAID0 RAID1 RAID3 RAID5 RAID6 RAID10
Data arrangement Dedicated parity Distributed parity

Striping Mirroring Double parity stripe Mirroring and striping
characteristics stripe stripe
Fault tolerance None High Medium Medium High High
Hot backup None Yes Yes Yes Yes Yes
Read performance High Low High High High Medium and high
Random write
High Low Medium Medium and high Medium and high Medium
performance
Sequential write
High Low Medium Medium and high Medium and high Medium and high
performance
mn ≥ 4
Number of disks ≥1 ≥2 ≥3 ≥3 ≥4 (m indicates the number of
RAID 0 groups.)
Available capacity All 1/n (n-1)/n (n-1)/n (n-2)/n 1/m
Erasure coding (EC )
Erasure coding (EC) is a data redundancy protection mechanism widely used in scale-out storage. Data written into
the scale-out storage system is divided into N data fragments, and M parity fragments are generated for the N data
fragments using EC. If M fragments are damaged in an EC group, the system implements data recovery from the N
fragments.
Compared with the multi-copy mechanism, EC improves both disk utilization and storage reliability, thereby cutting
costs. If N+M EC protection is used, the utilization rate of storage space is N/(N+M).
N* original data
N* original data fragments Any N* data
fragments fragments
Splitting Coding
File Original file
M* parity fragments
Contents
Stable Services, Represented by Production Transactions, Require Powerful
and Reliable Storage
Carrier: Core billing systems carry a Finance: Transaction channel diversity brings
heavier load in the 5G era sharp transaction volume increase
Online [The People's Bank of China

Outlet banking Report] Q3, 2019, banks
• Total number of accounts:
11.017 billion with 11.6%
Remote Telephone
banking YoY growth
Omni-channel banking
• Banks processed 59.464
transaction billion mobile payments
system with 31.5% YoY growth
Mobile Mobile
marketing banking
Mobile Network-wide Mobile phone
IoT devices IoT CDR
phone traffic CDR Emerging micropayment transactions increase
over 10x
User device growth: 3x CDR growth: 10x
24/7 service availability
Latency 5 ms -> 0.5 ms 99.999% -> 99.9999% Reliability
Agile Services, Represented by HPDA, Demand Efficient, Large-Capacity
Storage
Ultrasonic sensor Data collection and

AI training Simulation AI inference
preprocessing
GPS
Data
Data import AI training Simulation Verification
Lidar preprocessing
S3/NFS HDFS NFS NFS/CIFS NFS/HDFS
Millimeter-wave radar
(mmWave radar) Mass small files
100 GB/s Dozens of GB/s
100 PB level Ultimate OPS and < 1 ms latency
Camera bandwidth level bandwidth level
latency
L3 to L4 Each vehicle generates 64 TB of S3, NFS, and HDFS interfaces Requires high bandwidth and
50x more drive test mileage data daily. Massive data copy demands high OPS at once.
generates 50x more data. 100 PB data level requires efficient analysis. Ultra-low latency requires
more storage space. high performance.
Intelligent Management: Rapid Data Growth and Various Devices
Complicate Management
Digital transformation faces increasing challenges, including diversified
applications, media, and devices. Data volume managed per capita
Enterprise storage New mass data storage increases by several times.
(structured data) (unstructured data) Manpower
5
Required
CRM ERP EDA Container HPC Big data PB
Stream
1 PB Monthly deployment time
PB 15 days
required:
Billing Manufacturing Video AI Monthly deployment time
processing
required: 3 days Data Volume
O&M management
complexity increases.
Media Network Architecture
Centralized Distributed
architecture architecture
Planning Building O&M Optimization
Composable architecture Survey and Segment-by- Labor-
5 persons, 2 nights
Hyper-converged analysis segment check consuming
Manual
architecture 2 months 2 hours 3 hours
deployment
Data Security: With Digital Transformation in Full Swing, Data Protection
Becomes More Important than Ever
Software and hardware

Natural disasters Virus/Unauthorized access Human errors
failures
Fire in an ISP data center in Failure of core databases at Database of a listed
Ransomware in 2020
Europe a bank company deleted
>3.6 million websites down A large number of services > 150% increase in ransomware attacks Market value slashed by more than
Huge amounts of customer data interrupted for 37 hours 18 days of downtime on average caused by an HK$3 billion
unable to be restored attack CNY 120 million in compensation paid
UPS overload at an IDC service
Hacker attack on a bank to its clients
Datacom (Australia) hit by a provider
Code deleted by mistake
flood Power failure in the equipment room and Personal information of over 100
A large number of devices damaged, and system breakdown million users leaked Service downtime: 12 hours
user data permanently lost Services of multiple financial institutions Plummeting stock price and Direct loss: ~CNY 80 million
interrupted for 7 hours potential legal claims
Green Storage Is a Must to Achieve Its Carbon Peak and Carbon Neutrality
Goals
New energy methods to meet All-flash data centers to save energy and
carbon neutral targets reduce emissions
◆ Carbon neutral targets of major economies ◆ All-flash storage devices are becoming more ubiquitous in data
centers.
IDC Worldwide AFA ESS Market Forecast,
Value ($M), 2020-2025
According to
China EU US Japan
projections, the all-
Carbon peak Delivers the Rejoined the Announced the
flash market will grow
by 2030 Green Deal Paris Green Growth
Carbon Carbon Agreement Strategy by 9.6% to $14.8
neutrality by neutrality by Carbon Carbon neutrality by billion by 2022.
2060 2050 neutrality by 2050

2050
◆ Increased storage power consumption makes all-flash
◆ The key to carbon neutrality: New energy methods in devices essential to achieving energy goals.
According to IDC, China's data center storage Example: DC with 10 PB capacity
production and consumption areas, efficient energy use, and power consumption will grow rapidly in the next
energy conservation Energy five years.
Status quo After all-flash
Energy 21% CAGR
Carbon neutrality conservatio upgrade
upgrade in power
n consumption (in
current mode) 140,000 60,000
2.4x kWh/year
kWh/year
Production: HDDs
lower
Consumption: Efficiency Efficiency 20%: 2.4 TB SAS SSDs
Renewable reduction 100%: 7.68 TB SSD
Electric improvement 80%: 10 TB SATA
Emissions Absorption
Wind power, Electric vehicle (EV) Efficient Energy savings
photovoltaic Equal to the amount of CO2 absorbed by
and coal-to-electricity conversion, and waste
power, and 16,000 trees a year
and coal-to-gas storage, and reduction
hydropower projects use Report from Huadian Group:
Carbon emissions from 1 kWh electricity = 0.96 kg of CO2 = The amount absorbed by 0.2 trees a year
Thank you. 把数字世界带入每个人、每个家庭、
每个组织，构建万物互联的智能世界。
Bring digital to every person, home, and
organization for a fully connected,
intelligent world.
Copyright© 2023 Huawei Technologies Co., Ltd.

All Rights Reserved.
The information in this document may contain predictive

statements including, without limitation, statements regarding
the future financial and operating results, future product
portfolio, new technology, etc. There are a number of factors that
could cause actual results and developments to differ materially
from those expressed or implied in the predictive statements.
Therefore, such information is provided for reference purpose
only and constitutes neither an offer nor an acceptance. Huawei
may change the information at any time without notice.
SAN Storage
Contents
1. SAN Storage Overview
2. SAN Storage Networking
3. SAN Network Protocols
4. SAN Storage Technologies
5. SAN Storage Products and Application Scenarios
DAS (Direct Attached Storage)
Direct Attached Storage (DAS)
DAS is a dedicated digital storage device attached directly to a server or PC DAS is used to expand the capacity of a single
via a cable, as opposed to storage accessed over a computer network. A server.
typical DAS system is made of a data storage device (for example, JBOF)
connected directly to a host through a host bus adapter (HBA). Between Typical scenarios:
the host and storage, there is no network device (such as a hub, switch, or Storing cold data at a low cost, for example,
router). DAS can provide block-level storage services to hosts. backup, archiving, and video security.
DAS Types and Evolution
Built-in DAS External DAS Intelligent external DAS
Server Server Server

CPU CPU
CPU
RAM RAM
RAM
Disk Disk
Disk RAID
function HBA HBA
Controller
RAID
Just a Bunch of Disks (JBOD)
• Built-in hard disks of the server • The hard disks are placed in an external • Controller chips are added to the DAS to
• A limited number of disks deliver storage device. offload the RAID function to the DAS.
small capacity. • More but still a limited number of disks. • Provide simple data management
• Only capacity is provided. The RAID functions.
function is still provided by servers. • Another name is DAS-RAID storage.
Advantages and Disadvantages of DAS
Advantages
◼ Simplicity
DAS is connected to servers using cables for expansion, without the need for complex procedures, thereby
making it easy-to-use.
◼ Low cost
The price is more affordable than enterprise SAN or NAS.
Disadvantages
Following the data explosion and the new requirements in modern data centers, DAS has shown a lot of
disadvantages:
◼ Hard to scale Requirements of enterprise-
DAS can connect to one server (limited ports without network device, limited connected servers, and limited class storage:
connection distance).
◼ Inefficient utilization High scalability
DAS can be only accessed by a few servers and cannot share capacity or performance, resulting in data silos.
◼ Low reliability
High availability
As professional enterprise storage, DAS lacks disaster recovery (DR) features. High reliability
◼ Poor performance
Due to limited bandwidth and ports, the performance cannot be expanded on demand, consuming server Superior performance
computing resources. Easy to manage
◼ High management costs
Devices are managed one by one and cannot be centrally managed by management software. ...
DAS Storage Evolution
DAS SAN/NAS
Data silo Data silo Data silo • Improve scalability by network
Server Server Server devices. ... ...
• Can be shared among multiple
servers.
... ...
• Add rich DR features. FC/IP
• Achieve better performance. Switch Switch
• ...
JBOD JBOD JBOD
Storage device
Emerged in 1970s Emerged in 1990s
Data is scattered and not shared. Data is centralized and shared.
SAN Storage Overview
Definition:
Storage Area Network (SAN)
LAN bus 1. A network whose primary purpose is to transfer data
Application server File server Database server between computer systems and storage devices and
among storage devices.
A SAN consists of a communication infrastructure, which

provides physical connections, and a management layer,
which organizes the connections, storage devices, and
Data flow computer systems so that data transfer is secure and
robust. A SAN usually transfers block-level I/Os rather
FC/IP SAN network
than file access services.
Data flow
2. A storage system consisting of storage media, storage
devices, computer systems, and applications, plus all
control software, communicating over a network.
Storage device Storage device Storage device From: SNIA
Advantages of SAN over DAS
◼ Excellent flexibility
Contain plenty of disks, connect huge number of servers, and support scale-out controllers and scale-up disks to increase
performance and capacity linearly on demand, as well as supporting long-distance connections.
◼ Efficient utilization
Storage resources can be shared among multiple servers, eliminating data silos.
◼ High reliability
Various DR and backup features are provided to enhance reliability, such as replication, snapshot, and E2E DIF.
◼ High performance
High-speed and high-bandwidth network ports offload RAID calculation from servers to storage.
◼ Easy to manage
Provide centralized management and monitoring tools, reducing O&M OPEX.
SAN Storage Types
By Front-End Protocols By Architectures
Centralized SAN Distributed SAN

InfiniBand FCoE
NVMe over Fabric
IP FC
Comparison Among Different SAN Protocols
Type FC SAN IP SAN FCoE SAN IB SAN NVMe over Fabric SAN
Network NVMe over Fabric using RDMA;

Fiber Channel (FC) iSCSI FC over Ethernet (FCoE) InfiniBand
protocols NVMe over Fabric using FC
4 Gbit/s, 8 Gbit/s, 40 Gbit/s, 56

1 Gbit/s, 10 Gbit/s, 25 Gbit/s, 40 Gbit/s, 100 Gbit/s
16 Gbit/s/, 32 10 Gbit/s, 25 Gbit/s, 40 Gbit/s, 100
Bandwidth 25 Gbit/s, 40 Gbit/s, using RDMA, or
Gbit/s, or 128 Gbit/s, or 100 Gbit/s Gbit/s, or 200
or 100 Gbit/s 32 Gbit/s using FC
Gbit/s Gbit/s
Higher than iSCSI and
Speed High Low High Highest
slightly lower than FC
Higher than iSCSI and
Cost High Low High Highest
slightly lower than FC
From 1980s, the From about 2009, From about 2005, From 2017, providing high
From about 2001,
most common type, allowing the transfer of using in the performance for advanced
Others lower cost, around
around 75% market FC frames over Ethernet scenario requiring media such as all-flash and
20% market share.
share networks high bandwidth SCM.
Comparison Between Centralized SAN and Distributed SAN
Item Centralized SAN Distributed SAN
Hardware Dedicated: General:
architecture Uses dedicated hardware. Supports universal hardware architecture and x86 servers.
Closed source:
Open source:
Software Uses closed-source software, which is updated only by
Supports both closed-source and open-source software.
vendors.
Network FC/iSCSI/NVMe/IB iSCSI
Limited scalability: Better scalability:
Scalability Supports scale-up and scale-out expansion, but with limited Supports thousands of nodes, EB-level capacity, and on-demand capacity
number of nodes. expansion or reduction.
To be improved:
Higher reliability:
Compared with traditional enterprise storage, the evolving architecture
Reliability Mature architecture design, dedicated end-to-end design, and
uses universal x86 hardware that has lower reliability and less robust DR
various DR features
features.
Unlimited scalability but high latency:
Limited performance but low latency:
Unlimited peak performance and high bandwidth, but higher latency
Performance Limited maximum performance because of the limit in
because of the complex network interconnection and data inconsistency
controller quantity, but with lower latency
between distributed nodes
Application Mission-critical applications, which need high reliability and Cloud-based applications, which need elasticity and high bandwidth,
scenario low latency, such as databases such as hosting service in ISP
Comparisons Among DAS, SAN, and NAS
NAS
DAS SAN (Network Attached Storage)
Application server Application Application server

server
File system NFS/CIFS
File system
FC/IP switch IP switch
SCSI, FC
JBOD
RAID RAID RAID RAID
Protocol SCSI/FC/ATA FC/iSCSI TCP/IP

Transmission Type Block Block File
Typical Scenario JBOD for server Database, virtualization File sharing
Contents
SAN Storage Networking
I/O stack
Windows server Linux server AIX server Application
File system
Host layer ... ... Volume management
HBA HBA HBA Multipathing software
Hardware drive
HBA
Cables FC/IP Cables
Fabric layer SAN switch SAN switch Cable

Switch
Front-end interface module Front-end interface module
Storage layer Storage Storage

Storage Front-end interface module
controller
array array Storage cache
SAS/IP SAS/IP
Disk enclosure Disk
SAN Connectivity – Switch
Switch is the core component of a SAN, connecting servers and storage devices.
It allows for many-to-many automatic connections between senders and receivers, device name lookup, security, and
redundancy.
SAN Switch Selection Considerations:
What type? FC or iSCSI
How many ports? 48 or 96
How much bandwidth is required?
IP switch 16 Gbit/s or 32 Gbit/s FC, 10GE or 25GE IP
Physical partition or VLAN?
FC switch
Waterfall or mesh network?
IB switch
Broadcom
Mellanox Cisco
SAN Connectivity – HBA
The host bus adapter (HBA) is a circuit board or an independent adapter which can:
• Provide a physical connection between a host server and compatible devices/networks.
• Perform Input/Output (I/O) processing.
• Transfer data between the connected devices and the host computer.
• Free up resources of the host computer by conducting the data storage and retrieval operations on its own.
FC HBA iSCSI HBA IB HBA SFP

(small form-factor
pluggable transceiver)
SAN Connectivity – Cable
Optical Fiber Cable Ethernet Cable (RJ45) AOC/DAC
Transmission mode: Type: Active optical cable (AOC): optical modules

Single-mode fiber (SMF): used in long distance < CAT 5/5e/6 for 1 Gbit/s and optical fiber
10 km CAT 6e/7/7e for 10 Gbit/s Direct attached cable (DAC): copper cable
Multi-mode fiber (MMF): used in short distance
< 500 m Copper cable Transmission speed:
Transmission distance < 100 m 10GE/25GE/40GE/100GE
Interface type: Theoretical transmission distance: AOC ≤ 300
LC (for GE,10GE, 8/16/32 Gbit/s FC) m, DAC ≤ 10 m
MPO-PMO, MPO-4x LC (for 40GE/100GE)
Contents
SCSI Is the Basis of All
What is SCSI?
Small Computer System Interface (SCSI) is a set of standards for physical connection and data transfer between
computers and peripheral devices. The SCSI standards define commands, protocols, electrical, optical, and logical
interfaces. SCSI is most commonly used for hard disk drives and tape drives, but it can connect to a wide range of other
devices, including scanners and CD drives, although not all controllers can handle all devices.
SCSI is better and

faster than IDE
Item SCSI Integrated Drive Electronics (IDE)

Scenario Interface between computers and disks Interface between personal computers and disks
Performance 320 to 4000 Mbit/s 20 to 40 Mbit/s
Scalability Thousands of hard disks Two hard disks
Disk type Disks with over 10,000 rpm, and even SSDs Disks with over 5400 or 7200 rpm
SCSI Development History
SCSI-1
◼ Released in 1983–1985
SCSI-2
◼ Common SCSI
◼ 7-bit bus ◼ Adopted in 1988–1994
SCSI-1 ◼ Up to 7 devices ◼ Evolved from SCSI-1
◼ Standard instruction set
◼ 16-bit bus
◼ Up to 15 devices
SCSI-3
SCSI-2
◼ Approved in 1993
◼ Compatible with SCSI-1 and SCSI-2
SCSI-3 ◼ Supports both parallel and serial data
transmission.
◼ Supports multiple transmission protocols,
such as FC and iSCSI.
SCSI Transmission Model – Initiator and Target
Initiator Target
SCSI command (data block...) SCSI command (data block...)
Protocol (FC, iSCSI...) Protocol (FC, iSCSI...)
Physical interconnection (FC, Physical interconnection (FC,

ETH...) ETH...)
Cable
Server SAN storage

SAN Network Protocols
SCSI-3
Mapping layer FC-4 iSCSI FC-4 SRP
Transport layer FC-3 FC-3 IBA Operation

TCP
SAR
Network layer FC-2 FC-2
IP Network
Link layer FC-1 FCoE
Link
Physical layer FC-0 ETH Lossless ETH
PHY
FC iSCSI FCoE IB
Header
FC frame SOF (destination ID, source ID, frame type, Data CRC EOF
frame control flag...)
iSCSI frame
Ethernet header IP header TCP header iSCSI header Data Ethernet trailer
Fibre Channel Protocol
Fibre Channel (FC) is a transport layer protocol that is used to transfer data between computers and peripheral
devices. It has been primarily used for transporting SCSI packets from servers to SAN. FC was launched in 1988, and
acquired ANSI standard approval in 1994.
Advantages SCSI-3 Command
◼ Low latency
The lightweight protocol overhead of FC reduces FC-4 FC Protocol
transmission latency.
◼ High security FC-3 Common Fabric Services
The FC network is separated from the LAN for
higher security.
FC-2 Framing Protocol/Flow Control
Disadvantages
FC-1 Encode/Decode (8B/10B or 64B/66B)
◼ High costs
FC networks require dedicated adapters and 133 256 531 1 2 4 8 16 32
switches, and therefore are expensive to deploy. FC-0 Mbit/s Mbit/s Mbit/s Gbit/s Gbit/s Gbit/s Gbit/s Gbit/s Gbit/s
Obsolete data rates Current data rates

FC Network
Popular
Point-to-point Arbitrated loop Fabric
Support only two devices Support 126 devices Support up to 1,600,000 devices (FC
(Direct connecting) (FC hubs required) switches required)
iSCSI Protocol
Internet Small Computer System Interface (iSCSI) is a transport layer protocol that describes how SCSI data should be transported over a TCP/IP network,
providing an interoperable solution that can take advantage of existing Internet infrastructure. iSCSI was pioneered by IBM and Cisco in 1998 and
submitted as a draft standard by Internet Engineering Task Force (IETF) in March 2000.
Advantages
SCSI-3 Command
◼ Low costs
With standard Ethernet, the iSCSI protocol eliminates the iSCSI
need for expensive and complex switches and interface
modules like FC, making it more affordable.
◼ High flexibility TCP
iSCSI can run over IP, making it easy to connect and expand.
IP
Disadvantages
◼ Low reliability ETH
Standard Ethernet suffers from packet loss, so it is not as
reliable as FC. However, iSCSI is fit for general-purpose 10 25 40 100
applications. 1 Gbit/s
Gbit/s Gbit/s Gbit/s Gbit/s
FCoE Protocol
Fibre Channel over Ethernet (FCoE) is a storage protocol that enables FC Advantages
communications to run directly over Ethernet. FCoE makes it possible to
move FC traffic across existing high-speed Ethernet infrastructure and ◼ Simplified network structure
converges storage and IP protocols onto a single transmission cable and FCoE unifies input/output (I/O) interfaces and
interface. consolidates SAN and other traffic into the same
FCoE was approved by ANSI at 2009 and is rarely used today. network. This reduces the number of interface modules
and cables.
FC-4 FC-4
FC-3 FC-3 Disadvantages
FC-2 FC-2
◼ High reconstruction costs
FC-1 FCoE Mapping ◼ FCoE is designed based on lossless Ethernet.
Customers are reluctant to change or replace their
FC-0 ETH legacy networks with FCoE.
FC stack FCoE stack
IB Protocol
InfiniBand (IB) is a networking communications protocol that originated in
SCSI-3 Command
1999 and is used in high-performance computing that features high
throughput and low latency. It is utilized to establish data interconnections
between servers, between servers and storage devices, and between storage SRP
devices. Typically, it employs direct connection or switch networking modes.
IBA operation
Advantages SAR
◼ High speed
40 Gbit/s (QDR), 56 Gbit/s (FDR), 100 Gbit/s (EDR), 200 Gbit/s (HDR) Network
◼ Low latency
As IB reduces OS overhead, data can be transferred more quickly over the
Link
network.
Disadvantages PHY
◼ High costs
IB requires the dedicated adapter and switch. Generally, IB is more expensive 40 56 100 200
than FC. Gbit/s Gbit/s Gbit/s Gbit/s
NVMe-oF Protocol
What is NVMe? NVMe is the best alternative to SCSI.
Non-Volatile Memory Express (NVMe) is an interface
specification between computers and non-volatile
memory (especially SSDs). It was first proposed by the
NVMe-oF
NVM Express group in 2011.
(www.nvmexpress.org)
Type SCSI NVMe

NVMe-oF
SAN SAN NVM Express over Fabrics (NVMe-oF) is a transport protocol used
Scenarios
with HDDs/SAS SSDs with NVMe SSDs for communication between computers and peripheral devices
(especially all-flash storage). It allows the NVMe protocol to
Throughput Low High function on networks that use protocols like FC, RoCE, IB, and TCP.
Currently, the most commonly used protocols are NVMe-oF/RoCE
Latency High Low
and NVMe-oF/FC.
Queue Single Multiple
NVMe-oF/RoCE Protocol Stack
Software
NVMe command
RDMA software stack

RDMA
IB transport IB transport IB transport IB RoCE v1
Typical hardware
protocol protocol protocol
UDP
IB network layer IB network layer
IP
RDMA/Flexibility
IB link layer Ethernet link layer Ethernet link layer
IB + IP RoCE v2
InfiniBand RoCE v1 RoCE v2
InfiniBand Lossless Ethernet Lossless Ethernet
Now
NVMe-oF/RoCE Ecosystem
Category Component Status
✓Mellanox
NIC
✓Marvell
Servers NVMe-oF Driver ✓Linux: SUSE 15/15 SP1, Red Hat 8.0/8.1/8.2, Ubuntu 16.04.2 LTS…
✓Windows (third-party initiator required)
Multipathing software ✓VMware: VMware 7.0
✓Cisco
Network switch
Networks ✓Huawei
(support RoCE)
✓Mellanox
✓Huawei
SAN NVMe-oF/RoCE ✓NetApp
✓Pure Storage
Vendors are building compatibility with the NVMe-oF (FC) protocol.
Trend of NVMe-oF Protocol
Protocol Latency Scalability Cost
NVMe-oF (FC) ★★ ★★ ★
NVMe-oF (RoCE) ★★★ ★ ★★
NVMe-oF (TCP) ★ ★★★ ★★★
Huawei Storage NVMe-oF Roadmap

H1 2021 H2 2021 2022 2023
NVMe/FC
NVMe NVMe-oF/FC NVMe-oF/RoCE NVMe-oF/TCP
NVMe/RoCE
Replace Replace Replace
NVMe/TCP
SCSI FC-SCSI IB iSCSI
SAN Protocol Comparison – FC for Now, NVMe-oF for Tomorrow
Performance Cost Reliability,

Market Ease of Future
Availability, and
Share Mgmt. Upgrade Path
Protocol Latency Throughput Purchase O&M Flexibility
FC
iSCSI
FCoE
IB
NVMe-
oF/RoCE
Lowest rating Highest rating
In low-latency and high-reliability scenarios, FC is still the most popular protocol. However, NVMe-oF develops rapidly and will become
a mainstream protocol in high-throughput and high-flexibility scenarios.
Contents
SAN Storage Functions
Server
Value-added functions of LUNs
• Thin Provisioning Volume

• Tier
• Cache
Efficiency • QoS
improvement • Deduplication Switch
• Compression
Mapping
• Migration
• Virtualization
• Snapshot
• Clone
Data protection
• Replication LUN
SAN storage
• Metro
Introduction to LUNs
In SAN storage, a logical unit number (LUN) is the unique identifier assigned to identify a logical unit. The storage
system partitions physical disks into parts with logical addresses and allows hosts to read and write data in the
partitions. In this way, a partition becomes a LUN, which is a logical disk created on a SAN storage system.
Application
SAN storage LUN group Host group server
system
LUN 0
SAN storage
LUN 1 Mapping view
networking
LUN 2
SAN storage divides storage space into A host created on the storage system represents Once the application server has been connected
multiple LUNs based on service an application server. Storage space can be to the storage system, it can scan LUNs to locate
demands. LUN groups are created and allocated to application servers after a mapping the storage space allocated by the storage
mapped to host groups for better between a host group and a LUN group has been system.
management. established.
Introduction to Thin Provisioning
Logical virtual space
Thin Provisioning can dynamically
Server
allocate storage space on
Actual physical
demand when LUNs are in use, Storage
space
improving storage resource LUN
utilization. LUN
Pain points of traditional
allocation modes A LUN with the Thin Provisioning function is called
a thin LUN. After a thin LUN is created, the storage
• After being allocated to system does not immediately allocate physical
Mapping 1 2 3 4 5 6
specific LUNs, most of the storage space to the thin LUN. Instead, the storage
table system generates a mapping table to associate the
storage space remains D E F
logical address of the thin LUN with the physical
unused, resulting in wasted storage pool. In this way, space can be allocated
storage. dynamically based on servers' read and write
• Idle space is unavailable for D E F requests.
Thanks to thin LUNs, storage space can be allocated
new storage demands, so on demand, improving storage resource utilization.
space expansion is required,
incurring extra procurement Pool
costs.
Application Scenarios of Thin Provisioning
⚫ For core service systems that have demanding requirements on business continuity, such as
bank transaction systems, intelligent Thin Provisioning function helps expand system capacity
online without interrupting ongoing services.
⚫ For services where it is challenging to accurately evaluate the potential size of system data,
such as email and web disk services, this function can assist with on-demand physical space
allocation, preventing space wastage.
⚫ For mixed services that have diverse storage requirements, such as carriers' services, Thin
Provisioning can assist with physical space contention, achieving optimized space configuration.
Introduction to Tier
If the media performance varies in SAN storage, the Tier function supports dynamic data migration based
on data types (hot or cold) to achieve optimal global performance.
Performance tier (SSDs)
• High IOPS
Data migration policies:
• High capacity cost per terabyte Hot data migrated to the
Data with high access popularity performance tier • Time
Intelligent monitoring • Speed

Capacity tier (HDDs) • Bandwidth usage
• Low IOPS Cold data migrated to
• Low capacity cost per terabyte the capacity tier • Granularity
Data with large capacity but low access
popularity • ...
The Tier function is widely used and helps reduce procurement costs significantly in non-critical service
scenarios.
Introduction to Cache
The Cache function uses high-performance media storage space as the L2 cache between data disks and the cache to store read hot
data, improving the read cache hit ratio and read performance.
Generally, the L2 cache consists of

media that performs superior to data
disks. For example, SSDs and SCM
Deliver Result return Deliver write disks are usually used as the L2 cache
read I/Os upon a cache I/Os in hybrid flash storage and all-flash
hit storage, respectively.
RAM Cache RAM Cache
Applicable scenarios:
Miss • Read-intensive scenarios with
Read-hot data Read-hot data
random small I/Os, such as Oracle
SSD/SCM Cache stored in the L2 SSD/SCM Cache stored in the L2
OLTP databases and SQL databases
cache cache
Miss
Inapplicable scenarios:
SSD Tier SSD Tier Data flushing to • Write-intensive scenarios
preferentially write • Large I/O and sequential I/O
Miss scenarios
data into the capacity
tier
HDD Tier HDD Tier
Read I/O hit process Write I/O process
Introduction to QoS (Flow Control)
QoS is also known as intelligent service quality control. It dynamically allocates storage system resources to meet specific performance
goals of certain applications. With the I/O priority scheduling policy and I/O traffic control policy, QoS ensures uninterrupted key
services.
I/O priority scheduling policy
If QoS is disabled, the I/O processing of
Application 1
key applications may be delayed,
High I/O priority
resulting in high latency and poor
Application 2 service experience.
Medium I/O priority
If QoS is enabled, I/Os of mission-
Application 3 critical applications are preferentially
Low I/O priority processed, delivering optimal
QoS disabled QoS enabled performance experience.
I/O traffic control policy
I/O enqueuing
Implementing bandwidth limits for
I/O queue of Application 1/User 1; Bandwidth: ≤ 100 Mbit/s specific applications or users ensures
Processing reliable service performance and
... dequeued I/O
Disk
requests minimizes the impact on the overall
I/O queue of Application n/User n; Bandwidth ≤ 20 Mbit/s performance of storage systems.
Introduction to Deduplication
Deduplication is a data downsizing technology that deletes redundant data blocks in a storage system to save physical storage capacity,
meeting growing data storage needs. It includes inline and post-process deduplication.
Inline deduplication Post-process deduplication

Data to be Data to be
deduplicated deduplicated
Block 1 Block 2 Block 3 Block 1 Block 2 Block 3 Writes data onto disks
Write Similarity-based
deduplication opportunity
table
SFP 1 FP 1 SFP 2 FP 2 SFP 3 FP 3
FP* 1 FP 2 FP 3
Service handling during
off-peak hours
Comparison Fingerprint database Fingerprint database
Same FP with Same FP with Same FP with Same SFP

Different FPs Different FPs
different data same data same data (similar blocks)
Writes data onto Increases the Writes data onto Writes data onto Increases the Differential
disks reference count disks disks reference count compression *FP: Fingerprint
Application Scenarios of Deduplication
• Deduplication requires high performance in fingerprint query and random data read/writes. Therefore, this
function is typically only available for all-flash storage systems.
• Virtual desktop infrastructure (VDI) is a common example of data deduplication. VDI creates virtual images on a
storage device that contain a large amount of duplicate data, quickly causing a lack of storage space.
Deduplication removes the duplicate data between images to release storage resources for more service data.
Introduction to Compression
Data compression reduces occupied storage space by rearranging and compressing data without causing information loss, to improve
the efficiency of data storage, transfer, and processing.
• In essence, the data compression algorithm encodes duplicate data in a simpler

Source data
manner, to effectively reduce data size without causing information loss.
• Common data compression algorithms include Huffman Coding, LZ77, Rice, and
Run Length Encoding (RLE).
Data
compression • Data compression is often used together with deduplication for VDI, virtual server
infrastructure (VSI), databases, development and tests, and file service systems.
• Inline deduplication and compression reduce the amount of data to be written,

which helps prolong the lifespan of SSDs. By contrast, post-process deduplication
Stored on disks and compression require disk data to be further read and written, which may
shorten SSDs' lifespan.
Introduction to Migration
This function is critical to service data migration. It migrates all service data on a source LUN to a target LUN without interrupting host
services. The target LUN can then fully take over the source LUN to run services after the replication is complete.
• The data migration function aims to migrate data from one
Migration
physical space to another so that the host can read and write data
in the new space in the future.
1. Create a migration
• Synchronization of modified data ensures the host writes new data
pair.
correctly to the new space during data replication.
Host I/O A B C A B C
access D E F D E F • LUN information exchange enables the host to write data to the
2. Perform data new physical space with no impact on services after data
replication. migration is complete.
Source LOG
Target • This function is often used when upgrading storage systems, or to
H I J
LUN LUN improve service performance, modify LUN properties, and balance
3. Synchronize data
changes. data.
Source data volume Source data volume

LUN ID LUN ID
ID 4. Split the migration ID
pair.
0 0 1 1
0 1 LUN information 1 0
exchange
Introduction to LUN Virtualization
This function simplifies management of storage systems as it supports the local storage to take over services from a heterogeneous
LUN. It also supports data migration between heterogeneous LUNs without interrupting services.
• After the heterogeneous storage is connected to the local storage

by physical networking, the heterogeneous LUN is mapped
logically to the local storage as a logical space.
I/O
Virtualization • The local storage creates an external LUN using this logical space,
Application and maps this LUN to the application server, which can then
server External Logical Heterogeneous access the heterogeneous LUN through the local storage.
LUN space LUN
• LUN virtualization is mainly used to uniformly manage, migrate,
and protect data between heterogeneous storage systems. LUN
virtualization and migration work together to migration data from
heterogeneous storage system to local storage.
Physical
networking
Local storage Heterogeneous storage
Introduction to Snapshots
A snapshot is a point-in-time copy of source data that is accessible by the host and can serve as a data backup. Snapshots are mainly
used to protect production data.
A B C A B C A B C
Main features of snapshots
Source LUN D E F D M F D E F • Snapshots can be generated
within a few seconds to obtain a
G H I G H N G H I consistent copy of the source
J K L J K L J K L data.
• Because snapshots do not
Snapshot Unauthorized contain complete physical data
Recovery
creation modification copies, and instead serve as a
reference, they do not occupy a
A B C A B C A B C large amount of storage space.
D E F D E F D E F
Snapshot
G H I G H I G H I
J K L J K L J K L
20:00 21:00
Major Technologies of Snapshots
COW snapshot ROW snapshot
E New data
Snap 0 Snap 0 D Historical data

Snap 0 Snap 0
A B C D E A B C H I D E A B C D E A B C D E H I
20:00 21:00 Copying 20:00 21:00
• The Copy on Write (COW) technology uses the write-in-place mechanism to write new data to the original physical location. It also
copies the snapped historical data to another storage space, causing extra cost of data migration and reducing write performance.
• The Redirect on Write (ROW) technology uses the redirected-write mechanism, which writes new data directly and points the
metadata content to a new physical location. This ensures the snapped historical data does not need to be migrated when writing
new data.
• ROW facilitates high-density snapshots, snapshot cascading, and other technologies. It reduces the impact of snapshots on write
performance but its impact on continuity of data reads makes ROW more suitable for flash storage with high random read
performance.
Application Scenarios of Snapshots
⚫ Snapshot is one of the most commonly used technologies to protect service data.
⚫ In data backup systems, snapshots are often used for incremental backup.
⚫ Backup storage usually uses snapshots for data copy reuse.
⚫ Secure snapshots are used for ransomware protection.
Introduction to Data Cloning
Clones are a full copy of the source LUN data at a specific point in time. This copy is instantly accessible by the host and acts as a point-
in-time data backup.
Snapshot • The storage system can create a clone within a few seconds to
obtain a consistent copy of the source data. The clone can be read
A B C and written immediately.
2. Create a 3. Copy data to • Differential copies are supported to ensure data consistency
snapshot of the the target LUN. between the source and target LUNs before the clone pair is split.
source LUN. Clone pair splitting is also supported without service interruption.
1. Create a • Data cloning is often used to back up the source LUN data at a
A B C clone pair. A B C specific point in time, but also for data analysis of production data,
Source LUN Target LUN without impacting the production LUN.
Introduction to Remote Replication
Remote replication synchronously or asynchronously replicates data from the primary LUN at the primary site to the secondary LUN at
the DR site for remote data backup and DR. It is one of the core DR and backup technologies.
• Two LUNs form an active-standby replication pair, in which the

production site hosts the running services and the DR site is offline.
Production • When the production storage at data center A, the primary site, is
data faulty, services can be switched to data center B.
Replication pair
• There are synchronous and asynchronous remote replication
modes. The synchronous mode writes data to the primary and
Secondary secondary LUNs at the same time. The asynchronous mode
Primary LUN periodically synchronizes incremental data from the primary LUN
LUN
to the secondary.
Production data DR data center B

center A
Application Scenarios of Remote Replication
⚫ Remote replication is essential to DR and backup. All production storage systems should be configured using this
function to protect production services against data loss and avoid any serious impact on production and life.
⚫ The production storage and DR storage can be deployed in two data centers far away from each other, to provide
protection against data center-level disasters. They can also be deployed in the same data center to make the local
storage highly available.
⚫ Synchronous replication has a slight impact on production services. Typically, the active and standby data centers
should not be far from each other. There is no distance limit to asynchronous replication.
Introduction to Metro Clustering Function
This function allows two data centers to back up each other while they are both running. Unlike traditional DR services, if there are
faulty devices in one data center or if the data center is in faulty status, the other data center will automatically take over services. This
ensures high data reliability and service continuity, and improves utilization of the storage system resources.
• Real-time data synchronization is required between the two LUNs

of a metro pair. Both LUNs can provide services for application
servers at the same time.
• There is no primary/secondary relationship between the two data

Production
centers. Therefore, an intermediate arbiter is needed to determine
data
which data center continues to provide services when a fault
occurs.
Metro pair
• The metro clustering function is available in active-passive (A-P)
LUN A LUN B and active-active (A-A) deployment. In the A-P mode, only one of
the two storage systems provides services to the server and are
automatically switched-over to the other storage system if needed.
In the A-A mode, both storage systems can provide services to the
host at the same time. Most storage vendors support the A-P
Production data Quorum server Production data mode, only certain specialist vendors support A-A setups.
center A center B
Application Scenarios of Metro Clustering Function
⚫ The metro clustering function is also one of the most important DR and backup functions. To
protect core production services against interruption and loss, it is advised to configure this
function for all core production storage systems.
⚫ Active-active data centers can be separated at a distance of 50 km to 300 km, depending on the
application scope.
⚫ Active-active environments are common in finance, carrier, government, and healthcare.
Contents
SAN Storage Products
SAN storage products mainly comprise all-flash, hybrid flash, and scale-out storage.
All-flash storage Hybrid flash storage

Huawei New-Gen IBM
EMC PowerMax/ DellEMC FlashSystem
Power Store OceanStor UnityXT
Pure Storage NetApp NetApp FAS
FlashArray AFF HPE Primera
Huawei OceanStor
Dorado
Scale-out storage
Huawei OceanStor
H3C UniStor X10000
Pacific Red Hat Ceph
EMC ScaleIO
Typical Application Scenarios for SAN
Databases VSI VDI
Databases store business- Virtual server infrastructure Virtual desktop infrastructure

critical data and require the (VSI) often extends to ten (VDI) provides an enterprise
highest performance and thousand-scale virtual with large numbers of virtual
reliability specifications. machines (VMs) running a desktops and operating
broad range of operating environments, sometimes into
systems and applications, tens of thousands. By
with different performance centralizing the virtual
requirements. Reliability is a desktops, enterprises can
key issue because a failure manage data and protect data
can cause multiple application security more easily.
outages.
SAN Deployment for Databases
Key requirement
Data center 1 Data center 2
IP
SAN databases typically require minimally
IP
WAN 200,000 IOPS to process thousands of
Oracle RAC SQL Sysbase Oracle RAC SQL Sysbase transactions every second.
server 1 server server server 2 server server
◼ High reliability
Data center–level DR is required as
databases are often business-critical.
Active-active
SAN solution
◼ All-flash SAN storage
FC switches FC switches All-flash SAN storage provides extremely
high IOPS and low latency
(< 1 ms).
◼ Active-active/snapshot
IP
FC A-A setups provide excellent DR services
Mirror FC with zero RPO and RTO. In addition,
SAN storage SAN storage snapshots are configured to avoid logical
errors in data.
SAN Deployment for VSI
Data center 1 Data center 2 Key requirement
IP IP
WAN Each VM requires 20 to 50 IOPS. Therefore,
VMware FusionSphere VMware FusionSphere thousands of VMs require that SAN provides at least
100,000 IOPS.
◼ Data protection
Many applications are running on the SAN storage.
In the event of storage failure, data loss must not
exceed 30 minutes.
SAN solution
IP switches ◼ All-flash SAN storage

IP switches
All-flash SAN storage supports hundreds to
thousands of VMs.
◼ Remote replication/snapshot
Snapshots protect data from device logical errors
IP and virus attacks. In the event of SAN failure in one
Replication data center, remote replication allows services to be
link
Remote manually switched over to the other data center
SAN storage replication SAN storage with RPO no greater than 30 minutes.
SAN Deployment for VDI
Key requirement
◼ Performance
IP
Performance demands vary between virtual desktops and
applications. Typically, each desktop system requires 15 to 17
VMware Horizon
IOPS, and each data disk requires 3 IOPS.
Citrix
◼ High scalability
FusionAccess
Storage capacity and performance are required to expand on
demand as the number of virtual desktops increases.
SAN solution
IP switches ◼ Hybrid storage of SSDs and NL-SAS HDDs

A setup of SSDs and NL-SAS HDDs for system disks and data disks,
respectively, helps maximize cost effectiveness.
◼ Scale-up and scale-out
Scale up (vertical expansion) to expand the capacity by adding

disks, or scale out (horizontal expansion) to improve
SAN storage performance by adding controllers.
Thank you. Bring digital to every person, home, and
intelligent world.
Copyright©2023 Huawei Technologies Co., Ltd.


NAS Storage
Contents
1. NAS Overview
2. NAS Technology
3. NAS Products
4. NAS Applications
What Is NAS?
NAS, or network attached storage, is a framework that shares storage resources over a network and acts as a file
server for file access.
Features:
• Easy to use, no need for dedicated IT
experts
IP network
• Cost-effective, using IP switches
• Secure and reliable
RAID
• Easy data backup and recovery
NAS storage
History of NAS
The development of the Internet has created a need for unstructured data sharing, giving rise to the popularity of NAS
storage.
1946 - World's first 1974 - Invention of TCP/IP. 1984 - Sharing of network 1990 - The Internet connected
computer, but no 1979 - Duke University first implemented file servers achieved by IBM, Novell, commercial and enterprise
network. sharing without media. Microsoft, and 3Com. networks.
1946 1979 1984 1990
Media sharing File sharing File sharing NAS storage

FD, CD, HDD, USB and directory server
NAS Evolution
Huawei
Sun NetApp
Distributed NAS Startups Second-generation distributed NAS: Huawei
Microsystems Filer (first NAS
Isilon/GlusterFS WushanFS (renamed OceanStor 9000 OceanStor V5 NAS
NFS product)
in 2013)
1984 1988 1990 1993 2000 2006 2010 2011 2014 2018 2021
Huawei OceanStor Pacific next-

Apache gen scale-out storage (NAS and
IBM NetApp Huawei
Software Dell EMC lossless interoperability with S3
SMB (1996: NAS product: FAS OceanStor V3 and HDFS).
Foundation Acquired Isilon
Microsoft CIFS) Server 400 centralized NAS Huawei OceanStor V6 NAS.
HDFS
Differences Between DAS, SAN, and NAS
Item DAS SAN NAS
Reliability Network Direct connection Dedicated SAN LAN

Performance
Scalability Data sharing
Protocol SAS, ATA, SCSI FC, iSCSI, SCSI NFS, CIFS
RAID
Data package Block Block File
offload
File system FC HBA and iSCSI

HBA SAS HBA GE, 10GE
offload client
Data sharing Low High Highest
File sharing,
Small-scale Database and
Scenario archiving, and
servers VMware
backup
DR solution
Low High, dedicated High
(complexity)
Capacity Low High High
NAS Components
• NAS storage Client
• External server
a. DNS server
b. AD/LDAP server
• IP switch Switch
• NAS client
a. Server/Mainframe
b. Computer
AD
c. Printer
NAS
d. Scanner storage
External server
NAS Storage Software
• OS kernel
a. Windows NAS share (CIFS and NFS)
Web management
b. Linux
• Configuration
• Storage management
File system • View
a. RAID 0, 1, 10, 5, 6, 50
• Monitoring
b. EC (N+M)
c. Multi-copy • Statistics
Storage management • Troubleshooting
• File system RAID/Pool
a. OceanStor FS
b. Quota and WORM
• NAS share OS kernel (Windows and Linux)
a. CIFS, NFS, FTP, and HTTP
• Web management NAS hardware (CPU, cache, disk, and GE switch)
File System
File system: defines the data structure and management

for files stored on disks.
To enable data access on disks, a logical data storage
structure, such as a file system, must be established
between associated sectors. The process of creating a file
system on disks is called formatting.
Network File System
Application system Application system Application system Application system
File system client File system client File system client File system client
Network protocol
File system server

NFS and CIFS are common
network file systems for file
Storage sharing across heterogeneous
platforms.
Three Types of NAS Devices
Distributed NAS Centralized NAS Gateway NAS
... ... ...
IP switch
IP switch
NAS engine
...
Integrated storage and compute device
NAS storage SAN storage
Storage switch
NAS Type Device Model Scale-out RAID/EC Scenario Main Product Huawei
Integrated compute and storage
device (x86 server), no Dell EMC PowerScale
Distributed independent controller or
Up to 4000 EC (N+M)/Multi-copy Cloud and big data
Ceph
OceanStor Pacific
gateway
Controller + disk (capacity File sharing, archiving, and NetApp FAS/AFF OceanStor Dorado/OceanStor
Centralized expansion)
≤ 24 General RAID
backup Dell EMC Unity hybrid flash storage
File sharing, archiving, and HDS HNAS

Gateway NAS gateway + SAN storage ≤4 SAN-based RAID
backup IBM
OceanStor Dorado V3 (EOM)
Contents
1. NAS Overview
2. NAS Technology
3. NAS Products
4. NAS Applications
NAS Protocol Key Technologies
NFS SMB
NFS HDFS
Network File System (NFS) is a distributed file system protocol developed by Sun Microsystems (Sun) in 1984. It is
an open standard defined in a Request for Comments (RFC), which means anyone can implement the protocol.
In 2003, Sun Microsystems transferred NFS protocol development to the Internet Engineering Task Force (IETF).
NFSv4.1
NFSv4
NFSv3 5661
NFSv2 3010 + 3530
NFSv1 1813
1094
Version/RFC#
1980 1990 2000 2010
NFS SMB
NFS Evolution NFSv4.1 (RFC 5661) HDFS
NFSv4 (RFC 3530) 1. Enhanced agents
NFSv2 (RFC 1094) NFSv3 (RFC 1813) 1. TCP only a.Directory lease
1. UDP only 1. UDP and TCP 2. State mechanism b.Optimized reapplication and denial agent NFSv4.2
2. Stateless 2. 64-bit files 3. Client cache operations 1. Application I/O hints
3. 2 GB file read and 3. Asynchronous write (commit) 4. NFS server namespace c. File/directory notification mechanism 2. VAAI and ODX applications:
write 4. READDIRPLUS (virtual root directory) d.Clients' agent selection server offload read and write
4. NLM-dependent lock 5. More info in file properties to 5. Kerberos authentication 2. Session and multichannel mechanism 3. MAC: labeled NFS
mechanism prevent repetitive interaction 3. ACL enhancement 4. Space reservation
6. NFS ACL
6. Client authentication 7. File locking and share 4. Data retention 5. Sparse files
reservations 5. Notification of availability of byte-range locks 6. Application Data Block (ADB)
8. Replication and migration 6. Multi-server namespace (location, referral)
1989 1995 2003 2010 2012
Operating System Release Date/Period NFSv3 NFSv4 NFSv4.1

VMware ESXi 6.0–7.0 2015-09-10 Y N Y
Red Hat 4.0–6.3 2005–2012 Y Y N
Red Hat 6.4–8.3 2013–2020 Y Y Y
SUSE 9–11 SP2 2004–2012 Y Y N
SUSE 11 SP3–15 SP2 2013–2020 Y Y Y
CentOS 4–6.3 2005–2012 Y Y N
CentOS 6.4–8.3 2013–2020 Y Y Y
NFS SMB
NFS Working Principles HDFS
NFS implements remote communication based on the Remote Procedure Call (RPC) protocol. RPC uses the client-server
model.
Users and
1. The RPC client sends a call request with
applications
parameters to the RPC server and waits
File operation request and for a response.
reply NFS server 2. Upon receipt of the call request, the RPC
NFS client server obtains the process parameters,
outputs the calculation results, and sends
NFS request and reply the reply to the client.
3. The RPC client receives the reply and
RPC client RPC server
RPC request and obtains the call results.
reply
NFS SMB
NFS Version Differences HDFS
NFSv3 NFSv4 NFSv4.1

NFSv3 client NFSv4 client NFSv4 client
Sessionless concurrent Session-secure serial Session-secure parallel

short connections persistent connections persistent connections
Security difference Performance difference

Item NFSv3 NFSv4 NFSv4.1
User Host-based local Kerberos Kerberos
Small files Large files
authentication authentication authentication authentication NFSv3 NFSv3
IP address, IP address segment, host name, or network group NFSv4 NFSv4
Client control
name
File permission NFSv4.1 NFSv4.1
UGO permission ACL permission ACL permission
control
NFS SMB
SMB Evolution HDFS
• Server Message Block (SMB) is a protocol for network file sharing. One of the most popular versions is Microsoft SMB.
• The Common Internet File System (CIFS) Protocol is a dialect of SMB. Both SMB and CIFS are also available on ESXi, Unix, Linux and
Mac.
MS WIN WIN
DOS XP VISTA WIN7 WIN 8.1 WIN 10
1988 1992 1996 2000 2001 2003 2007 2008 2009 2012 2013 2016
Linux WIN WIN WIN WIN WIN

IBM Samba Server Server
SMB1.0 2000 2003 2008 2012 2016
Operating System Release Period SMB1.X SMB2.X SMB3.X

Windows XP, Windows 2000, Windows Server 2003 2000–2003 Y N N
Windows Vista, Windows Server 2008, Windows 7 2007–2009 Y Y N
Windows Server 2012, Windows 8.1, Windows 10, Windows
2012–2016 Y Y Y
Server 2016
NFS SMB
NFS vs. CIFS/SMB HDFS
Item NFS CIFS/SMB
Accessing Operating System Linux, Unix Windows
Development Group IETF Microsoft
Security Authentication Client IP, domain user Local user, domain user
Supported Domain System NIS, LDAP AD
Session State Stateless Stateful
Transport Protocol TCP, UDP TCP
NFS SMB
HDFS HDFS
HDFS (Hadoop Distributed File System) was designed and developed based on a Google File System (GFS) paper. In
addition to the features of other distributed file systems, HDFS provides:
• High error tolerance: Hardware is considered always unreliable.

• High throughput: High throughput support is provided for applications that have massive amounts of data access.
• Large file storage: Data storage at the TB or PB level is supported.
HDFS is suitable for: HDFS is not suitable for:

Random writing of a large number of small files
Large file storage and stream data access
and low-latency read
…
Who is using HDFS
NFS SMB
Hadoop HDFS Architecture HDFS
Hadoop HDFS component

HDFS mainly works in active/standby mode, with its architecture consisting of three
components: name node, data node, and client.
⚫ Name node
Name node File
▪ Stores and generates metadata for a file system
1 ▪ Runs one instance
2
3
4 ⚫ Data node
▪ Stores the actual data and reports blocks it manages to the name node
▪ Runs multiple instances
⚫ Client
▪ Supports service access to HDFS and obtains data from the name and data nodes and
1 2 1 3 sends it to services
2 1 4 2
4 3 3 4 ▪ Runs multiple instances together with services
Data node
Hadoop file management
⚫ A file is split into blocks (default size: 64 MB), and each block has multiple
Client copies stored on different machines. The number of copies can be
specified (default: 3) when the file is being generated. This ensures data
reliability.
⚫ A file cannot be modified after being created, written, or closed.
NFS SMB
Quotas HDFS
Problem: With rapid growth in information assets

and file sharing, storage space management is Quota 1: 1 TB
NAS
becoming more and more complex.
Manager
Share: 5 TB
When multiple users access a shared directory, Quota 2: 1 TB
some users will overuse the space and others will
not even be able to use it. In the worst cases, the Dev team
entire system will run abnormally.
Quota 3: 1 TB
Solution: Quota mechanism
Test team
By limiting the file capacity or number of users,
users can be prevented from occupying excessive Quota 4: 1 TB
storage resources, thereby improving system
reliability. Others QA
NFS SMB
Quota Levels HDFS
Three Quota Levels Example
Level Threshold I/O Restriction NAS

Advisory Soft Soft
Denying I/O operations and reporting

Hard Quota High Directory 1
alarms
90% Soft 700 GB 850 GB
fs/0
Reporting alarms and allowing data
writes during a grace period, restricting
Soft Quota Middle Directory 2
data writes immediately after the
90% Soft 1000 files 3000 files
expiration
Only reporting alarms and not restricting Directory N

Advisory Quota Low
writes 90% Soft 1 TB 2 TB
NFS SMB
Quota Working Principles HDFS
Quota Support Matrix

Dimensions
• Capacity Resource Level Directory User User Group
• File quantity
Advisory quota Y Y Y
Capacity Soft quota Y Y Y

Objects
• Directory Hard quota Y Y Y
• User Advisory quota Y Y Y

• User group
File quantity Soft quota Y Y Y
Hard quota Y Y Y
NFS SMB
WORM HDFS
Write Once Read Many (WORM) is a data Differences between WORM files and
protection mode. After data is written, the file common files
enters the protection mode through manual setting Operation WORM File Common File
or after a certain period of time.
Read
Modify
What applications support WORM?
Delete
CD/DVD-ROM, electronic exam, electronic contract,

Rename
and archive
Others?
NFS SMB
WORM Modes HDFS
• Enterprise WORM: Allows administrators to flexibly

Mode Differences
manage files. This mode is mainly used for internal
Enterprise Compliance
Operation
WORM WORM
enterprise control.
Privileged
deletion
• Compliance WORM: Enables enterprises to protect
SEC 17a-4
data in compliance with laws and regulations, so to compliance
prevent legal risks when archiving confidential
documents.
NFS SMB
Global Namespace HDFS
Solution: Global Namespace

NAS Nightmare
(GNS)
Users and
Users and
applications
applications • File virtualization: Aggregates file
systems and provides unified
Global namespace
namespace.
• GNS allows clients to access files even if

they do not know the location of discrete
files, similar to accessing a website
without knowing its IP address.
File systems
File systems
NFS SMB
DNS HDFS
• DNS
The domain name system (DNS) is a network service that Domain name resolver
translates domain names into IP addresses.
• DNS Server Functions

✓ Domain name resolver
✓ Load balancing
• Benefits
✓ Access to the Internet without having to remember
each IP address
✓ More balanced access, no single-point bottleneck
NFS SMB
DNS-based Load Balancing HDFS
Client
DNS server • Principles
1. A user uses a domain name to access NAS services.
2. The DNS client sends a DNS request to the DNS server to
obtain an IP address based on the domain name.
Frontend 3. The DNS server selects an IP address and returns it to the
service client.
network
• Load balancing policies
NAS storage
1. Round-robin
2. CPU usage of each node
3. Number of connections of each node
4. Port bandwidth usage of each node
5. Comprehensive load of each node
Node 1 Node 2 Node 3 Node 4
NFS SMB
Embedded and External DNS Servers HDFS
Embedded DNS topology
• External DNS server options include Client
Windows DNS server and Open BIND.
Advantages: unified management of multiple

NAS systems
• Embedded DNS server

NAS/DNS server
Advantages: high reliability, low cost, and simple
networking
NFS SMB
NAS Audit Logs HDFS
NAS audit logs are used in security audit scenarios to trace each file operation. When a file is accessed, the system
records the operation in NAS audit logs.
• File operations:
a. Create, delete, and rename
User A User B
b. Open and close
c. Read and write
d. Get attributes (Get_attr) and set attributes
• Operation time (Set_attr)
• Operator Syslog server
• Operation type Transmission e. Get security attributes (Get_security) and set
Audit logs
security attributes (Set_security)
• Supports integration with third-party log

Storage system
servers.
NFS SMB
Multi-Tenancy HDFS
Pain points
• Security issues arise when enterprises or users use the same physical storage device and access or interfere with each other's
logical resources.
• IT service providers need to pay extra costs to manage users.
Each tenant has independent NAS protocol services,

including:
1. Domain service (AD, LDAP, and NIS)
2. CIFS service
Tenant A Tenant B Tenant C 3. NFS service
CIFS & NFS CIFS & NFS CIFS & NFS
4. NDMP service
Each service can be disabled and enabled separately.
Physical storage system
NFS SMB
Multi-Protocol Access HDFS
Multi-protocol access enables Windows, Linux, and Unix clients to access the same directory or file at the same
time.
Multi-protocol access security
SMB client NFS client Item Permission User

Local user
Security SMB NT ACL
AD collaboration AD server
authentication
SMB 1, 2, 3
File lock
UNIX mode Client
NFS
SMB 2, 3, 4 NFSv4 ACL NIS/LDAP
SMB-NFS multi- LDAP, NIS
DNS, AD
protocol access
Consistency of shared access files
Item Reading File Writing File
DNS query, AD
Reading file Yes No
NIS, LDAP
File lock Writing file No No
NAS storage
NFS SMB
Multi-Protocol File Sharing HDFS
Application scenario: enterprise office file sharing

Enterprise office Enterprise office
LDAP, NIS server
Management
Name server AD server Anti-virus server

Performance Service Sharing
User management
monitoring management management
Contents
1. NAS Overview
2. NAS Technology
3. NAS Products
4. NAS Applications
Centralized NAS: OceanStor Dorado
OceanStor Dorado all-flash unified storage sets a new benchmark with its industry-leading stability, SAN and NAS performance, intelligence, and efficient
Overview
management and O&M.
• OceanStor Dorado 3000 is an entry-level storage product, boasting high cost-effectiveness to expand its footprint in the market.
• OceanStor Dorado 5000 and OceanStor Dorado 6000 are mid-range storage systems that outperform competitors with their high quality and high density.
• OceanStor Dorado 8000 is an entry-level high-end storage system. Its stable architecture and competitive mid-range pricing are key to expanding its market share.
Products • OceanStor Dorado 18000 is a high-end storage system. Its stable architecture, top-notch performance, and intelligence are key qualities that allow it to expand its
presence in high-end markets and empower benchmark projects for core NAs.
NAS Focusing on high-performance NAS scenarios, such as EDA simulation, carrier CDRs, and financial data exchange platforms
Competitiveness in high-end markets

Rock-solid architecture + leading intelligent
technologies
Unified SAN and NAS storage

Entry-level all-flash High quality + high-density models
storage with a Mid-range all-flash storage
minimum initial
Recommended OceanStor Dorado 8000 OceanStor Dorado 18000
configuration of 128
configurations GB
• Tolerating failure of 3 out • High-end architecture
of 4 controllers • NVMe/intelligent technologies
OceanStor Dorado 5000 OceanStor Dorado 6000 • Full mesh for core applications
OceanStor Dorado 3000
Scale-Out NAS: OceanStor Pacific Series
High-performance Content resource pool Log retention Smart city Virtualization/Cloud
resource pool Database
computing Backup &
Application Web disk
archiving
layer
NFS/SMB
S3 HDFS SCSI/iSCSI
/POSIX/MPI-IO
File storage service Object storage service HDFS storage service Block storage service
SmartMulti- Self-encrypting drive

SmartTier SmartQuota SmartQoS SmartEqualizer SmartThin SmartQoS
Tenant (SED)
Software SmartEncryption SmartAuditlog SmartInterworking Recycle bin SmartIndexing MultiPool SmartDedupe SmartCompression
layer
HyperReplication
Object Versioning SmartTakeover HyperSnap HyperReplication (A) HyperSnap HyperClone HyperMetro
(S)/(A)
EC/Multi-copy Fast reconstruction SSD cache DIF
OceanStor OceanStor OceanStor

OceanStor OceanStor Pacific 9350 OceanStor OceanStor Pacific 9550 Pacific 9950
Products Pacific 9340 Pacific 9346 OceanStor Pacific 9540 Pacific 9546 OceanStor
Pacific 9920
Pacific 9520
Contents
1. NAS Overview
2. NAS Technology
3. NAS Products
4. NAS Applications
Energy Exploration
Data collection Data Data
Result processing Data archiving
and import preprocessing analysis/processing
Seismic data collection Seismic data processing Seismic data interpretation/reservoir simulation
Reservoir Backup and
Regular Pre-stack Post-stack Seismic data Well data simulation & archiving
Service Data import processing migration modification interpretation interpretation imaging
process Result data
Drilling and logging data
Service I/O model: GB-level files, sequential large I/Os Image loading: 100,000 IOPS
Data size: 1–20 TB/day
requirements Large bandwidth: 2–20 GB/s (per PB) Latency: ms-level response
• Seismic data processing + interpretation/reservoir simulation: OceanStor Pacific 9920/9950 + OceanStor

Solutions
Pacific 9546/9550 + automatic tiering
Gene Sequencing
Service process Data
Data collection High-performance computing (HPC) HPC-AI/big data Data archiving
preprocessing
Sample Library AI-/big-data-assisted Data archiving
preparation Format Sequence Mutation diagnosis and treatment
preparation
conversion alignment analysis Permanent retention 5-year retention
0.5 hours 4.5 hours
4.5 hours 12–72 hours 2–6 hours
BCL format FASTQ file BAM/SAM file VCF file AI model XXX file
Service requirements
Archived data amount
AI training &
100 GB/sample inference
10 PB/device
Intermediate data amount Low cost and automatic
1 PB/year/device Single thread: 6+ GB/s Strong demands for
5 PB/year/device management
Compressed genetic data transfer file systems
Bioinformatic analysis (sequencing analysis + report Post-gene sequencing Data archiving

Library preparation (sample upload to sequencers
interpretation) Clinical and scientific Mass genetic and
+ data download to storage)
File format conversion, decompression, gene splicing, alignment, research assistance is genomic data is
A sequencer downloads data to its storage, which is archived at a low
sorting, deduplication, mutation detection, and combined genotype provided based on AI and
configured to 1 PB on average. cost.
detection are included. big data.
Solutions • Production and archive storage: OceanStor Pacific 9546/9550
Autonomous Driving
Service process and requirements
DT data Data Data preprocessing AI training X-in-the-loop Backup/archiving

injection verification/simulation
Radar MIL simulation
Import
.bin files Data detection/cleaning, labeling Cleaned data
into the SIL simulation
server Model training
Weather HIL simulation
.log files 64 TB/day/vehicle
Large write bandwidth: tens Multi-protocol access, large High IOPS, ultra-low Read bandwidth: tens to
Video/image of GB/s read bandwidth latency of < 1 ms hundreds of GB/s
Files
NFS/SMB/S3 NFS/HDFS NFS/S3 NFS/S3 NFS/S3
GPS
.log files
Scenario library Backup/
DT data Training set Result data archiving
• Scenarios involving AI training: OceanStor Pacific 9920/9950 + OceanStor Pacific 9546/9550 + automatic
Solutions tiering
• Common scenarios: OceanStor Pacific 9546/9550
Non-Linear Media Editing System
Production Long-term
Ingesting system Editing system
system archiving
Data Data Data

migration migration migration
Functions and requirements for production Functions and requirements for

1. Stable high bandwidth, large files, large I/O blocks (> 1 archiving
MB) 1. Massive capacity
Solutions 2. SMB 2/SMB 3 in Windows and macOS clients 2. Low TCO
3. Easy scalability for performance and capacity
4. Easy data migration
Solution: OceanStor Pacific 9920/9950, 9546/9550 Solution: OceanStor Pacific 9546/9550
EDA System
User
NAS applications
Client (VM) Computing (physical)
EDA • EDA development
software • EDA test
Cadence Service features
Task Short (< 3 hours)
submission Synopsys
Mentor • Mass small files: > 4 billion
Task management
and scheduling (IBM LSF) • File size (95% of files) < 128 KB
Result return Normal (< 24 hours) License • Concurrent access, high OPS, and low
obtaining latency
Task
Active Standby scheduling Solution: OceanStor Dorado
License server
Long (> 24 hours)
NAS
Red Hat Linux CentOS
NFS 10GE
NAS storage
resources
Archiving
Medical PACS System
Doctors' terminals Image capture Technicians
Medical X-ray MR/MRI
Clinical Outpatient
treatment
process UCG EMG ECG
PACS servers DICOM server PACS DB server PACS app server
PACS image NAS storage

storage
PACS functions and requirements Solutions

• Mass small files (most ranging from 128 KB to 1 MB) • Primary storage: OceanStor Dorado/Hybrid flash storage NAS
• High OPS and low I/O latency • Archive storage: OceanStor Pacific series
• High reliability
Video Security System
Video security architecture

Functions and requirements
• Large files and I/O blocks (> 1 MB)
Management platform
Video storage • High-bandwidth sequential write
Smart application • Massive capacity, easy scale-out
• Centralized management
NAS
storage
Direct storage/dump
Solutions
• Dump: OceanStor Pacific 93xx
• Direct storage: IVS3800
Big Data Analysis
Real-time Operational Data

Data mining … Unified
retrieval analysis visualization portal
Application Functions and key requirements
layer • On-demand computing and storage
expansion to avoid resource wastage
• Compatible with big data clusters on the
live network to protect existing
Loader investments
Spark Storm Flink
MapReduce Elk HBase • Stringent requirements for cost
Computing reduction due to mass data sets
YARN
layer
Solutions
• OceanStor Pacific 9546/9550 big data
Native HDFS storage
Storage layer
HDFS storage
intelligent world.


Object Storage
Security level:
Contents
1. Object Storage Overview
2. Object Storage Features
3. Object Storage Products
4. Object Storage Application Scenarios
Storage Trends: Unstructured Data Surge
Users: > 800 million
Douyin Daily active users: > 600 million
Monthly active users: > 1.9 billion
Tik Tok Avg. video searches per day: > 400
180 ZB > 80% million
New data in 2025 Unstructured data
• There has been an explosion in users, videos, and capacity.
Unstructured • Videos require quick storage and retrieval.
data • Videos require tagging, categorization, and analytics.
• User preferences are created from usage history for
Structured data customized video recommendations.
2005
2005年 2010
2010年 2015
2015年 2020
2020年 2025
2025年 • The most popular topics will be updated dynamically.
20%
• Data is accessed through mobile apps.
IDC report: In 2025, 80% of the projected global volume of new

data, totaling 180 ZB, will be unstructured data. Traditional NAS storage provides inadequate performance,
▪ Unstructured data: Image, video, audio, radar, and sensor data scalability, and retrieval. ByteDance has developed object
▪ User data analytics: 35% of data storage for Douyin videos.
Object Storage Harnesses Internet and Mass Data
• Limitless scalability
✓ Storage of tens of billions of objects
Unstructured data surge ✓ Linear expansion on demand for
capacity and performance
• Storage must provide large capacity
and scalability on demand to handle
sudden data growth. • Customized
• Unstructured data must have attributes
metadata for tagging, classification, ✓ Identified by unique
object IDs Object
and analytics. Object
✓ User-defined attribute
fields
Interactive online access
• Universal access
• Data centers must support access across ✓ Online device access
Object storage
regions. anytime and anywhere
• File systems (NAS) are designed for
human-machine interaction, meaning they
struggle to handle M2M interactions, such
as data sharing and management. • Flexible connection
✓ RESTful API, HTTP, HTTPS
Object Storage Surpasses NAS for Mass Data
NAS storage Object storage
Server
Shared directory Direct visit
Access
App App • Servers • Support for different
• App access through devices
shared directories Networks IoT Mobile
Directory
Protocols
CIFS/NFS
HTTP/HTTPS
• Complex protocols HTTP/HTTPS (RESTful API)
NFS/SMB/POSIX • Incompatibility with
• URL address
• Simple flexibility
different OSs
Network
Private Public/Private
LAN Internet
• Private connection • Public/private connection
• Within LANs • Unrestricted networks
Bucket Bucket Bucket

Tree structure Flat structure
Organization
• Retrieval • Simple file
performance of directories
files or directories • Better scalability Object
impacted by growth
NAS storage: Local file directory Object storage: Function
What Is Object Storage?
Object storage simplifies management and reduces costs by eliminating hierarchies for unlimited
cloud storage of unstructured data.
Object identifier
Data Object System

metadata
User
Bucket Bucket Bucket metadata
Item Description
Object identifier
Object name
(OID)
• Data is stored as objects.
Object attributes, such as object size, access permissions, and
• Data is contained in a flat address space. System metadata
creation time
• Bucket/Object
• Easy scalability User metadata Additional object information about applications
Data Service data such as text, images, or video
Objects with Data and Metadata
User metadata File (Data)

▪ X-ray
System metadata (object attributes)

System metadata
▪ File name: CA12345.JPG
▪ Created: June 2, 2009
▪ Last modified: July 4, 2009
User metadata (user-defined attributes)

File (Data)
▪ Name: XXXXXX
▪ Category: Chest X-ray
▪ Save time: XX years
▪ Doctor: Dr. XX
▪ Result: No abnormalities found
Object
History of Object Storage
• IBM acquired Cleversafe and
launched object storage COS.
NetApp acquired Bycast • The Ceph community
(an object storage startup) Inktank released
EMC launched content- HDS acquired ArC released a basic commercial
and renamed it Ceph enterprise Huawei launched
addressed storage (CAS) software from Archivas version.
StorageGRID. edition. FusionStorage 7.0
to manage mass data and and renamed it HCP
object storage
Centera (a CAS-based object storage in 2009.
supporting S3.
product) for archiving.
1996 2002 2006 2007 2008 2010 2012 2013 2014 2015 2016 2019 2021
Carnegie Mellon
University (USA) Amazon released AWS (S3)
proposed object EMC released Atmos
object storage, which Huawei launched Huawei launched
storage as a S3, which is mainly
officially introduced object OceanStor 9000 OceanStor Pacific, the
research project. used in cloud Huawei launched its
storage as a cloud storage V300R005 file next-generation scale-
The University of services and first self-developed EMC released ECS,
service and introduced it storage, supporting out storage system
California, Berkeley, backup/archiving object storage solution supporting S3 and
into the cloud computing NFS, S3, and Swift. supporting S3 and
then launched a scenarios. — OceanStor UDS — Swift, which still
field, setting an industry lossless interworking
similar program. supporting EB-level receives iterative
standard for Internet access with NAS and HDFS.
scalability. upgrades today.
to object storage.
Distributed Architecture vs. Gateway Architecture
The distributed architecture provides object storage
without additional gateways.
⚫ Distributed storage
Gateway architecture Distributed architecture Object mapping for file systems is eliminated to
shorten access latency.
Object gateway
⚫ Concurrent access
Object APIs Distributed storage improves over object gateways
and NAS architecture by making the network fully
Protocol
conversion Object APIs redundant and eliminating single points of failure
for better concurrent access.
File system
⚫ Scale-out
Capacity and performance increase linearly with
nodes, which are unrestricted by file system size.
• DELL EMC PowerScale • DELL EMC ECS
• IBM COS • Huawei OceanStor Pacific
• NetApp StorageGRID • Red Hat Ceph
API Types
AWS S3 OpenStack SNIA
⚫ Amazon S3 is an open object ⚫ Swift is an object storage API in ⚫ CDMI is a cloud storage API
storage protocol of Amazon AWS the OpenStack cloud computing released by the Storage
and is the de facto standard of framework. Networking Industry Association
object storage. (SNIA).
Amazon S3 sets the mainstream standard for object storage APIs
AWS S3: A Standard Protocol for Object Storage
Account
User 1
• Amazon S3 is an open protocol launched by User 2
Amazon AWS for object storage.

• Industry standard for object storage.
Bucket Bucket
• It simplifies the complex directory structure of the
file system with the bucket/object model.
Object ... Object Object ... Object
• Uses HTTP/HTTPS RESTful APIs.
Concept Description
Bucket A bucket is a container for storing objects, eliminating the nested structure and container hierarchies.
Object Object data and metadata are always stored in buckets.
AWS S3 APIs: Request and Response Instances
Operation Format Instances
Setting the region of a bucket
PUT / HTTP/1.1 Request HTTP/1.1 200 OK Response
User-Agent: Jakarta Commons-HttpClient/3.1 Server: OBS
Host: example.obs.example.com
x-obs-request-id: 001B21A61C6C00000134029F41D1527F
Accept: */*
x-obs-id-2:
Date: Sat, 03 Dec 2011 07:31:58 +0000
PUT MDAxQjIxQTYxQzZDMDAwMDAxMzQwMjlGNDFEMTUyN0ZB
Authorization: OBS QUFBQUFBQWJiYmJiYmJi
BF6C09F302931425E9A7:QBaO+tS/76QYHVnUoxvf9EPH/3o=
object Location: /example
Content-Length: 149
<CreateBucketConfiguration xmlns="http://obs.example.com/doc/2015-06- Date: Sat, 03 Dec 2011 07:31:58 GMT
30/"> Content-Length: 0
<Location>china</Location>
</CreateBucketConfiguration>
Obtaining the region of the bucket

GET /?location HTTP/1.1 Request TTP/1.1 200 OK Response
Authorization: OBS Server: OBS
04RZT432N80TGDF2Y2G2:JUtd9kkJFjbKbkP9f6T/tAxozYY= x-obs-request-id: 367CB63A2F283044981285492719060
GET Date: Sun, 26 Sep 2010 09:16:00 GMT x-obs-id-2:
MzY3Q0I2M0EyRjI4MzA0NDk4MTI4NTQ5MjcxOTA2MEFBQUFBQUFBYmJiYmJiYm
User-Agent: Jakarta Commons-HttpClient/3.1
object JD
Host: example.obs.example.com Content-Type: application/xml
Date: Sun, 26 Sep 2010 9:18:36 AM GMT
Content-Length: 560
<?xml version="1.0" encoding="UTF-8" standalone="yes"?>
<Locationxmlns="http://obs.myobsdomain.com/doc/2015-06-30/">cn-north-
4</Location>
OpenStack Swift: Similar to AWS S3
Account
User 1
⚫ Swift is a distributed storage module of OpenStack.
User 2
⚫ The module consists of an account, a container (also a buck in
AWS S3), and an object layer.
⚫ Uses HTTP/HTTPS RESTful APIs. Container Container
... ...
Object Object Object Object
Description
The top layer of the hierarchy. The service provider creates an account and grants ownership of the resource to the user. The account
Account defines namespaces for containers. Containers in different accounts are allowed to have the same name.
Container Create namespaces for objects. Objects in two different containers can share the same name, but they represent two different objects.
Object Storage content, such as documents, images, and customized metadata.
Contents
Data System
Data Access
Protection Management
Important Concepts
• Bucket: A bucket is a container for storing objects. It can be
Account considered as a directory that can be accessed through the
network. Buckets cannot be nested, thereby avoiding complex
Network
multi-layer structure.
User User • Object: Objects represent data and are regarded as files that can
be accessed through the network. An object is always stored in a
bucket.
Bucket Bucket Bucket
Object
Object Object
Object
Object • Account: An account should be obtained before using the object
Object
Object Object Object storage service (compatible with Amazon S3 APIs), and acts as the
owner of resources.
• User: A user is created by an account and belongs to the account.

An account can control how users utilize resources by granting
users different permissions.
Data System
Data Access
Object Storage API: Requires Software or Tools

Object storage APIs need to use commercial software, customer-developed software, or other tools on the market, such as the
S3 Browser or backup software.
S3
Bucket
Bucket
Object
Object
Service host Object
S3 Object
APP
2. Customer-developed
API Bucket
software (API)
SDK (C, Python, S3 Object Object System
Object
Java, PHP) Key
metadata
Object User-defined
metadata
Data
S3
Common client tools:

• Cloud web disk tools: Cloudberry Object storage system
and Onebox
• S3 Browser
Data System
Data Access
GNS: A Unified Domain Name Access Storage

An object storage system includes multiple regions and AZs. Storage resources in different regions need to be virtualized and
aggregated into a global namespace for domain name resolution, location services, and load balancing.
www.Huawei.com
A unified domain
name
Region 1 Region 2 Region N

LAN/WAN
AZ 1 AZ 2 AZ N
Object
Object
Cluster 1 Cluster 2 Cluster 3 Cluster N Object storage Object
Object
Object resource pool Object
Node 2 Node 2 Node 2 Node 2 City A City B City C City D

...... ...... ...... ......
Node N Node N Node N Node N
www.Huawei.com/1 www.Huawei.com/2 www.Huawei.com/3 www.Huawei.com/4
Data System
Data Access
Erasure Coding: System-level Data Redundancy Protection
• Erasure Coding (EC): It is a data encoding and fault tolerance

technique of object storage. It replaces the multiple-copy technology
Principle with data redundancy verification, achieving similar or even higher
reliability, much higher disk utilization, and lower costs.
• Redundancy protection mode: It is usually known as the N+M mode,

where N is the data fragment number and M is the redundancy parity
Data fragments
fragment number.
Parity ✓ When data is written, the system divides it into N data fragments, and
PP PP PP PP
fragments calculates M parity fragments through an encoding algorithm.
✓ When a data fault occurs, the system reads any N fragments (regardless of
data fragments or parity fragments) to recover the other M fragments.
4+2 EC mode
• Redundancy protection level: Unlike the RAID disk-level redundancy

… protection of traditional centralized storage, EC supports higher node-level
and cabinet-level protection. When M nodes or cabinets become invalid, EC
can ensure service continuity and zero data loss.
Data System
Data Access
Versioning
The object uploaded for the third time
after versioning is enabled • Function
Versioning is used to restore an object that is overwritten or

Key=exp.jpg The latest-version object
Key=exp.jpg
ID=33 deleted mistakenly by storing multiple versions of an object in the
Key=exp.jpg The object uploaded for the second time same bucket. Versioning allows users to save, query, and restore
ID=22 after versioning is enabled
objects of different versions, so lost data can be easily restored
Key=exp.jpg The object uploaded for the first time
Enabling ID=11 after versioning is enabled when misoperations or program faults occur. Versioning can also
versioning
Key=exp.jpg An object uploaded before be used for retaining and archiving data.
ID=NULL versioning is enabled
Disabled • Status
It indicates whether the versioning function is enabled or disabled

for a bucket. The versioning function of most product buckets is
Enabled Paused
disabled by default, and the status can be configured and queried.
Different statuses
Data System
Data Access
Multi-Tenancy: Isolates Data Stored by Tenants

The combined use of QoS policies and multi-tenancy that isolates tenants ensures that the specified performance level and
capacity are always available under changing conditions. The data of each tenant is isolated and invisible to other tenants.
Tenant A Tenant B Tenant C

• A single system: Provides services for multiple
clients, reducing initial investments.
• Online storage: Data is centrally managed by a

Bucket
A
Bucket
B
Bucket single system and is logically isolated.
C
• Transmission encryption: Supports HTTPS

encryption and user authentication to ensure data
Object storage transmission security.
Data Center
Data System
Data Access
Data Encryption Protects Object Data from Leakage
Host Object app Major functions
⚫ Implements data-at-rest encryption based on software.

⚫ Supports flexible data encryption by tenant and namespace.
⚫ Supports the international standard algorithms AES128-XTS and
S3 AES256-XTS.
KMS Internal key

manager Encryption/Decryption Key Manager Service (KMS)
External AES128/AES256...
Unified key ⚫ Supports external and internal key managers.
key Encryption engine
manager ✓ External key manager: complies with the KMIP and supports
manager
KMIP 1.0, KMIP 1.1, and KMIP 1.2.
✓ Internal key manager: generates AKs as secure random numbers,
Storage pool adopts a three-layer architecture to protect key security, and

supports key backup, restoration, update, and destruction.
Object storage
Data System
Data Access
Cross-Site DR
Cross-site DR type Features
⚫ Multi-active DR ⚫ Multiple sites carry services at the same time, tolerating site-level
✓ Multi-active replication: Multiple sites work in active-active mode. faults.
Each site has one copy of data. ⚫ The distance between sites is long — measured in thousands of
✓ Cross-site EC: Multiple sites work in active-active mode. One data kilometers.
slice is distributed across multiple sites, and multiple sites share one ⚫ A global namespace allows any site to access the same object.
copy of data. ⚫ Different DR types can be set for different buckets. Multi-active, multi-
⚫ Remote replication: Sites work in active-passive mode, supporting active- site EC, and remote replication can coexist to meet different SLA
standby switch. Each site has one copy of data. requirements.
9-site EC
Triple-active Active-active
Data System
Data Access
Quota
Object storage uses the quota function to limit the
resources occupied by buckets and tenants.
Object storage • Quota
100 TB ✓ Bucket-level quota: specifies the maximum capacity of a bucket. When

the bucket size reaches the specified upper limit, new data cannot be
written into the bucket.
✓ Tenant-level quota: specifies the maximum capacity of an account. When
the total size of buckets of an account reaches the specified upper limit,
Tenant 1 Tenant 2 Tenant 3 the account cannot write new data.
General Financial Personnel
20 TB manager dept. 40 TB dept.
40 TB • Resource statistics
✓ Bucket resource statistics: include the bucket size and number of objects
in the bucket. Users can query their own bucket resources.
✓ Account resource statistics: include the account quota, number of
buckets and objects owned by the account, and the total capacity.
Employee A Employee B Employee C Employee D Account resources can be queried.
2 TB 10 TB 8 TB 5 TB
• Notification
Bucket 1 Bucket 2 Bucket 3 Bucket 4 ✓ An alarm or event is triggered when resources are occupied.
Data System
Data Access
API Billing by Operation or Capacity
• Object storage billing records the operations of end users and the size
of data stored by end users, and generates service detail records.
Billing center
✓ Billing services are mandatory for object storage and you only pay for what you
use.
✓ Object storage generally provides open APIs, such as SFTP, which upper-layer
billing uses to access object storage and obtain data for fee calculation.
API Billing APIs • There are two types of bills:

✓ Detailed records save the operations of end users, such as object
Object storage
downloads/uploads or bucket creation/deletion. Recorded attributes include
the operation type and data capacity.
✓ Inbound bills record the resource utilization of end users. Object storage saves
a capacity record for each uploaded object in the inbound bill file.
Contents
Huawei OceanStor Pacific Series Builds an Optimal Storage Foundation
for Diversified Mass Data
High-performance Content resource pool Virtualization/
Database
computing Backup & Log retention Smart city Cloud resource pool
Web disk
Application archiving
layer
NFS/SMB
/POSIX/MPI-IO S3 HDFS SCSI/iSCSI
Self-encrypting drive
SmartTier SmartQuota SmartQoS SmartEqualizer SmartMulti-Tenant SmartThin SmartQoS
(SED)
layer
HyperReplication
(S)/(A)
EC/Multi-copy Fast reconstruction SSD cache DIF ...
OceanStor OceanStor OceanStor OceanStor

OceanStor OceanStor Pacific 9350 OceanStor Pacific 9550 Pacific 9950
Products Pacific 9340 Pacific 9346 OceanStor
Pacific 9540 Pacific 9546 OceanStor
Pacific 9920
Pacific 9520
Contents
Backup and Archiving of Massive Data Improve Data Recoverability with
Lower Storage Costs
Production Office
Online data Scenario description
• Backup: Object storage replaces tape libraries to retain data for a long time,
providing high availability and improving data recoverability.
• Archiving: Object storage saves important but rarely accessed data, such as
Backup/Archive design, regulatory, and financial documents. For data with low usage
data Backup/Archive data frequency, it is important to achieve a balance between cost-effectiveness and
availability.
Application 1 Backup media server
Storage requirements
Production Office
• Security: WORM, audit logs, and encryption
Online data
• Availability: tolerance for site-level faults
Object storage • Cost: high space utilization and massive scalability
Backup/Archive
data Backup/Archive
data
Solution
• OceanStor Pacific 9550/9540
Application 2 Backup media server
Content Storage Is Oriented to Massive Data, Replacing NAS to Improve
Access Efficiency
Scenario description
Email Check image Traffic checkpoint Media • The enterprise management system for electronic content is oriented to large-scale
digital content directory services, such as the PACS system in the healthcare industry
and check images and dual-recording in the financial industry. The data types are
documents, images, and videos.
• Based on the service log workflow, the system comprehensively manages
Video unstructured data and provides search and file management functions.
Music VOD surveillance Storage leasing
• The number of files increases from tens of millions to billions.
Storage requirements
• Performance and high scalability
Multi-tenancy ✓ Billions of objects, second-level latency
✓ Metadata retrieval: It takes less than 1s to retrieve billions of objects.
✓ QoS guarantee
• Availability: active-active, tolerating site-level faults
Object storage
Solution
• OceanStor Pacific 9550/9546
intelligent world.


portfolio, new technology, etc. There are a number of factors that could
cause actual results and developments to differ materially from those
expressed or implied in the predictive statements. Therefore, such
information is provided for reference purpose only and constitutes
neither an offer nor an acceptance. Huawei may change the information
at any time without notice.
Backup and DR
Security level:
Contents
1. Introduction to Business Continuity
2. Introduction to Backup Solution
3. Introduction to Disaster Recovery Solution
Why Is Business Continuity Important?
Consider the following example. Assume as a senior executive you need to handle the following incidents:
1. Trading data is damaged due to human error (securities).
2. All insurance policies are lost due to a power failure (insurance).
3. A year's worth of geological is lost due to malicious actors (oil exploration).
4. Medical records are unusable due to damage to the tape library (healthcare).
...
All of these incidents impair business, including data loss, downtime, and impacts to brand image. What is
the solution?
The Necessity of DR – Direct Benefits
Cost of downtime/hour
Apple Google
Amazon iCloud storage Full breakdown
Two breakdowns breakdown Global traffic
The amount of The number of decrease in 5
Loss ($) affected users minutes
7,000,000 3,000,000 40% 648
280
Source: Huawei MI 200
160
63 110
From conventional DCs to cloud environments, downtime resulting from a 9
disaster (natural, human-error, equipment failure, or cyber threat) can cause (Unit: US$10,000)
financial and reputational repercussions. Customers require 24/7 business Media Healthcare Retail Manufacturing Telecom Energy Finance
continuity and data protection. Source: Network Computing, the Meta Group and Contingency
Planning Research
Key Parameters Used to Measure Business Continuity – RPO and RTO
The recovery point objective (RPO) is the amount of data loss The recovery time objective (RTO) is the maximum length of a time it
that an enterprise can tolerate losing. This threshold is typically should take to restore systems to normal conditions in order to avoid
represented by a time value. further or unacceptable consequences caused by downtime.
Available zone in production Available zone in DR DC Available zone

Available zone in
DC in DR DC
production DC
The amount of data App service Failover
loss time
Two Major Standards of DR Construction
RPO: the amount of data loss caused by downtime RTO: the time period of downtime
Information Security Technology – Technical Requirements of
IBM's SHARE 78 Seven Levels of Disaster
Disaster Recovery Specifications for RPO RTO the Classified Protection of
Recovery (International)
Information Systems (China) Cybersecurity 2.0 (China)
Tier 6 – Zero data loss and services are Tier 7 – Zero data loss and services are Level 4
0 < 15 min Local backup + remote
automatically started automatically started
backup + local high
Tier 5 – Zero data loss Tier 6 – Zero data loss 0–30 min < 2 hours availability (HA) +
remote service HA
Tier 5 – Real-time status update in the DR DC 2–12 hours < 24 hours Level 3
Tier 4 – Electronic transmission and full device Local backup + remote
support real-time backup + local
Tier 4 – Active DR DC 2–24 hours < 24 hours service HA
Tier 3 – Electronic transmission and partial

Tier 3 – Electronic vaulting 12–24 hours 24 hours
device support Level 2
Local backup + remote
24 hours to several 24 hours to several periodic backup
Tier 2 – Backup site support Tier 2 – PTAM + hot site
days days
Tier 1 – PTAM Several days Several days

Level 1
Tier 1 – Basic support Local backup
Tier 0 – No off-site data Several days Several days
Differences Between Backup, DR, and Archiving
 Backup:  DR:
✓ Recovery of lost and damaged data ✓ Recovery to the nearest running time
✓ Retrieval of historical data
Backup VS. DR ✓ Direct service takeover
✓ No direct service takeover ✓ Unable to recover lost or damaged data
Archiving
 Archiving:
✓ Only retains data copies while deletes
raw data
✓ Long-term information retention
✓ Meets compliance and cost-efficiency
requirements
Backup provides data recovery, DR focuses on business continuity, and archiving meets
regulatory compliance.
Contents
Basic Concepts of Backup
Definition:
A backup is an additional copy of production data used to recover lost or damaged data. With ever-increasing data
volumes, organizations need an efficient and cost-effective backup method to meet service and regulatory
requirements on data storage, retention, and availability.
DB Disk
OS
Disc
App
Tape library
Backup
File
Cloud
And more
And more
Components of a Backup System
Production DC Backup DC
LAN
Backup management system

… ◼Backup management software and server.
A A A A ◼Automatically initiates data backup, including
reading data from backup clients, and writing
App server DB server File server App server Media Backup
data to the backup media.
server management
server
SAN
Backup media
Production storage
Backup network Backup media

Backup client ◼Devices storing backups.
◼Provides a transmission path of backup data streams from the
◼The service host to be backed up needs to ◼Common backup media: magnetic disk storage, cloud
backup client to the backup server.
run a backup software agent. storage, tape library, virtual tape library (VTL), CD-ROM
◼Generally consists of TCP/IP network or FC.
tower, and more.
⚫ Centralized backup solution: The backup media servers (running backup software) are separated from the backup media, that is, storage and
computing are decoupled.
⚫ All-in-one backup solution: integrates backup software, media servers, and backup media, also known as coupled storage.
Data Backup Process
When data backup is initiated, important network communication takes place between the components of the backup
infrastructure. This operation is initiated either by the backup management server or the client.
1 The backup management server starts the periodic backup.

App server/Backup client
The backup management server retrieves backup-related
2 information from the backup directory.
The backup management server notifies the media server to
File DB server Mail 3 load backup media.
server server
The backup management server notifies the backup client and the
4 latter sends the data that needs to be backed up to the storage
node.
4 5 The backup client sends the data to the storage node and updates
5 the backup directory on the backup management server.
1 3 6 6 The media server sends data to the backup device.
2 7 The media server sends metadata and backup information to

8 the backup management server.
Backup management 7 Media Backup device 8 The backup management server updates the backup directory.
server server
Data Recovery Process
Once receiving a recovery request, the administrator can find the list of backed up clients in the recovery application.
The backup client sends a data recovery request to the

App server/Backup client 1 backup management server.
2 The backup management server scans the backup directory to

identify the data to be recovered and the target clients that will
File DB server Mail
server server receive the data.
3 The backup management server notifies the media server to load

backup media on the backup device.
1
4
4 The media server reads the data and sends it to the backup
client.
3 4 5 The media server sends the recovery metadata to the

backup management server.
2
6 6 The backup management server updates the backup directory.
Backup management
5
Media server Backup device
server
Three Typical Scenarios of the Backup Solution
Local backup Copy archiving Copy replication
Production DC Production DC Production DC DR DC
Software copy
… … …
replication
Software copy
tiering
DB File server Backup DB File server DB File
Backup Backup Backup
server server server server server
server server server
IP/FC IP/FC Copy tiering IP/FC
Storage copy
replication
OceanProtect OceanProtect OceanProtect OceanProtect
Production storage Backup storage Production storage Backup storage Archiving storage Production storage Backup media Backup media
Category Local Backup Copy Archiving Copy Replication

Data in the production system is replicated and The backup system replicates local copies to
Solution The backup system tiers copies and transmits them from level-1 backup
transmitted to the backup system over a network, with the DR DC, implementing remote
Description storage to level-2 archiving storage.
multiple historical copies retained. redundancy of multi-copy data.
Remote data DR, multiple copies in multiple
Hot and cold data tiering, long-term data retention, and compliance of policies
Main Feature Local data backup and copy management DCs, and compliance of policies and
and regulations
regulations
Backup Software/Storage-level replication
Direct application backup and software backup Software tiering and storage tiering
Technology technology
High-performance and high-deduplication ratio featured High-performance and high-deduplication ratio featured storage media and
Backup Media Homogeneous software or storage
storage media large-capacity and low-cost archiving media
Common Backup Types
Synthetic full backup
Full backup Differential backup Incremental backup
(Forever incremental backup)
Sun.
Mon.
Tue.
Wed.
Thu.
Fri.
Sat.
Sun.
✓Full backup every day ✓ Full backup once a week

✓ Full backup once a week ✓ One-time full backup
✓ Daily backup of the
✓ Daily backup of changes ✓ Daily backup of the
change compared with
from the last full backup in change compared with
last backup in the rest of
Recommended the rest of the week last backup
the week
scenarios
✓ Core applications ✓ Important businesses ✓ Important businesses ✓ General businesses
✓ Small data volume (less ✓ Medium data volume (less ✓ Medium data volume (less ✓ Large amount of data (around
than 1 TB) than dozens of TB) than dozens of TB) hundreds of TB or even PB) with
hundreds of millions of small
files
Typical Backup Networking
LAN-Based LAN-Free Server-Free
Production DC Backup DC Production DC Backup DC Production DC Backup DC
LAN Dedicated backup network
… … …
iDA iDA iDA iDA MA CS iDA iDA iDA iDA MA CS iDA iDA iDA iDA MA CS
App File App Media Backup App DB server File App Media Backup App DB server File App Media Backup
DB server
server server server server server server server server server server server server server server server
IP/FC IP/FC IP/FC
Production storage Backup media Production storage Backup media Production storage Backup media
Description Description Description

 Backup data streams are transmitted over the  Backup data streams are transmitted over a SAN  Backup data streams are transmitted directly
existing LAN. or a dedicated backup network. through the media server over an independent
SAN.
➢ Pros: ➢ Pros: ➢ Pros:

 Makes full use of the existing network,  Minimal impact on the existing service  Minimal impact on service hosts
reducing investment network  Minimal impact on the existing service network
➢ Cons:  High backup performance  High backup performance, depending on the SAN
 Occupies the existing service network ➢ Cons: ➢ Cons:
bandwidth  Increased investment in the network  Increased investment in the network
 Limited backup performance  High requirements on backup devices  High requirements on backup devices
 Impact on host applications  Impact on host applications
Overview of Huawei Backup Solution
All-in-one backup
Intra-city backup DC Remote backup DC
Replication
DPA OceanProtect DPA OceanProtect

Production system
Centralized backup Huawei

AWS
Intra-city backup DC Remote backup DC
File DB Public cloud
Backup Software Backup Software

Mail VM Replication
OceanProtect backup storage OceanProtect backup storage

OceanStor data storage OceanStor data storage
Contents
What Is Disaster Recovery?
Disaster recovery (DR) is the coordinated plan to restore and recover systems, data, and the infrastructure from
an event that disrupts business operations. It is the process of restoring a data copy and applying logs or other
necessary processes to that copy to bring it to a point of consistency. After all recovery efforts are completed, the
data is validated to ensure that it is correct.
Power failure Viruses outbreak Fire Device failure
Major Disaster Recovery Technologies
Application Application
Database mgmt. system Database mgmt. system

Host layer
OS OS File system
File system • Application replication
• Database replication
Raw device/volume Raw device/volume
• Logical volume replication
Device I/O driver Device I/O driver
IP
SAN-based network SAN IP
SAN
layer
Huawei
Solution
Disk array layer
Disk array NAS Disk array NAS
DR Implementation in the Industry
Data replication is implemented on hosts without compatibility issues for underlying devices.
Pros During database replication, the DR center takes over certain workload from the production center.
Host layer: Typical replication
software includes Oracle Data
Guard, Veritas VVR, DSG, and Database replication can run on only single-type databases.
Quest Cons Host-layer replication occupies certain host resources, affecting the application system.
Host-based data replication is complex and usually requires system reconstruction.
Broad compatibility enables integration of different backend heterogeneous SAN storage resources.
Network layer: Typical Simultaneous disaster recovery for multiple SAN arrays without one-to-one mapping.
Pros Extendable disaster recovery platform.
gateway devices include EMC
No extra investment required even with additional hosts and arrays.
VPLEX and IBM SVC
High initial investment and few solution vendors.
Cons Complex gateway networking increases latency.
Data replication does not affect the host application system and the DR architecture is easy to deploy.
Storage array layer: Arrays that

Pros If the production arrays fail, applications can be switched to the DR arrays with minimal downtime.
Data replication runs on the underlying array, eliminating fees incurred by host licenses.
support dual-write or mirroring,
such as Huawei storage series Storage arrays at both sites must be provided by the same vendor, with only certain vendors (such as Huawei)
providing heterogeneous solutions.
Cons Certain solutions don't support real-time remote data access (Huawei SAN active-active architecture is an
exception), requiring data volume improvement or snapshots.
Overview of Huawei Flash Storage Disaster Recovery Solution
Local High Availability Active-Active Data Center Geo-Redundant 4DC DR Solution
Active-Passive DR Solution Geo-Redundant 3DC DR Solution
Solution Solution (SAN Storage 6.1.6)
• Financial core production systems
• Financial core production systems (class
Service application (class A and B) • Financial core production systems
A and B) • Financial core production systems (class A and B)
• HIS and PACS (classes A, B, C, and D)
• HIS and PACS • HIS • Financial core production systems (class A and B)
• Core system of the carrier's BSS • HIS and PACS
• Core system of the carrier's BSS domain • Core system of the carrier's BSS domain • Core manufacturing system
domain • Core system of the carrier's BSS domain
• Core manufacturing system • Core manufacturing system
• Core manufacturing system • Core manufacturing system
• University information system
• University information system
• Single point of failure (SPOF) of • Quick recovery in the event of power
• Failure of apps, storage devices, or an • Rapid recovery in the event of large-scale or regional • Rapid recovery in the event of large-scale or regional
components (network adapter, outage, fire, or flood
entire data center disasters such as earthquakes, fires, and floods disasters such as earthquakes, fires, and floods
Fault protection controller, and storage device) in a • Planned migration in the event of
• Zero service disruption • Planned migration in the event of planned power • Planned migration in the event of planned power outages,
data center planned power outages, routine O&M,
• Zero data loss outages, routine O&M, and disasters routine O&M, and disasters
• SPOFs of networks in data centers and disasters
VM DB VM FS
FS DB VM FS DB VM FS VM DB VM FS DB FS
DB VM DB VM FS DB VM FS DB FS
DB VM DB VM DB VM DB VM
Solution topology
Local A-A A-A Sync/Async A-A/Sync
Sync/Async
Production storage DR storage A-A Remote async A-A
Production storage DR storage Production storage DR storage Production storage DR storage Remote storage
Production DR storage Remote production Remote DR
storage storage storage
Distance (from Intra-city or remote data center Intra-city or remote data center (synchronous
Different cabinets in the same Intra-city active-active (< 100 km); remote asynchronous
production to DR equipment room or at the same site
Intra-city (< 100 km) (synchronous replication: < 100 km; replication: < 100 km; asynchronous replication: < 3000
replication (< 3000 km)
storage) asynchronous replication: < 3000 km) km)
0 (synchronous) 0 (active-active/synchronous) 0 (intra-city active-active)

RPO 0 (local active-active) 0 (intra-city active-active)
minutes (asynchronous) minutes (asynchronous) minutes (remote)
≈ 0 (intra-city)
RTO ≈0 ≈0 Minutes Minutes or hours
minutes (remote)
Storage + Optical
Connection
N/A Implemented in Version 6.1.6 Implemented in Version 6.1.6 Implemented in Version 6.1.6 Implemented in Version 6.1.6
Coordination
(SOCC)
Deployment cost Low High Medium High High
Overview of Huawei Scale-Out Storage Disaster Recovery Solution
HyperMetro Active-Active Data HyperReplicationSynchronous HyperReplication Asynchronous HyperGeoEC Geo-Redundant Multi-DC

HyperGeoMetro Geo-Redundant Multi-DC Solution
Center Solution Replication–Based DR Solution Replication-Based DR Solution Solution
(Object Storage)
(Scale-Out Block Storage) (Scale-Out Block Storage) (Block/File/Object/HDFS) (Object Storage)
• Financial document image & dual-recording
• Financial database system • Financial database system system
Service application • Financial VM system • Financial VM system • Financial database & VM system • Financial document image system
• Financial log data archiving system
• Financial new scale-out core system • Financial new scale-out core system • Manufacturing content (email, etc.) management • Financial dual-recording system
• Scientific research data archiving system
• Government cloud • Government cloud system • Financial data exchange system
• Government tax data archiving system
• Smart city • Smart city • Government data archiving system • Manufacturing content (email, etc.) management system
• Unstructured data (email, etc.) archiving system
• Public safety policing cloud and policing big • Public safety policing cloud and policing big • Financial new scale-out core system • MSP enterprise 2B service
data data • Smart city
• Public safety policing cloud and policing big data
• Failure of applications, storage devices, or an • Quick service recovery in the event of power • Quick service recovery in the event of power • Rapid recovery in the event of large-scale or regional disasters • Rapid recovery in the event of large-scale or regional
entire data center failure, fire, or flood failure, fire, or flood such as earthquakes, fires, and floods disasters such as earthquakes, fires, and floods
Fault protection • Zero service disruption • Planned migration in the event of planned • Planned migration in the event of planned power • Planned migration in the event of planned power outages, • Planned migration in the event of planned power outages,
• Zero data loss power outages, routine O&M, and disasters outages, routine O&M, and disasters routine O&M, and disasters routine O&M, and disasters
S3 app S3 app S3 app S3 app S3 app S3 app

DB VM DB VM DB VM DB DB VM
DB VM VM S3 app S3 app
S3 app S3 app S3 app S3 app
Solution topology
A-A Sync Async Async Async Async replication Async
replication replication replication
Production storage DR storage Production storage DR storage
Production storage DR storage
DC 1 DC 2 DC 3 DC 1 DC 2 DC 3
Async replication
Async replication
Distance (from
production to DR Intra-city (< 100 km) Intra-city (< 300 km) Asynchronous replication (< 3000 km) Asynchronous replication (< 3000 km) Remote asynchronous replication (< 3000 km)
storage)
> 0 (seconds, depending on cross-site network bandwidth, latency, > 0 (seconds, depending on cross-site network bandwidth,
RPO 0 (intra-city active-active) 0 Seconds to minutes
and frontend service pressure) latency, and frontend service pressure)
RTO ≈0 Minutes Minutes Minutes (depending on load balancer) Minutes (depending on load balancer)
Deployment cost Medium Medium Medium High Low
Synchronous/Asynchronous Remote Replication and Active-Active Arrays
Synchronous Remote Replication Asynchronous Remote Replication Active-Active Arrays

DC 1 DC 2 DC 1 DC 2 DC 1 DC 2
Host 1 Host 2 Host 1 Host 2 Host 1 Host 2

Synchronous Asynchronous
remote replication remote Metro cluster
replication
Storage 1 Storage 2 Storage 1 Storage 2 Storage 1 Storage 2
Synchronous Remote Replication Asynchronous Remote Replication Active-Active Array

Architecture A-P A-P A-A
RPO RPO = 0 RPO > 0 RPO = 0
RTO RTO > 0 RTO > 0 RTO ≈ 0
Distance < 100 km < 3000 km < 100 km
≥ Peak-hour bandwidth (at least 64 ≥ Changed data volume during peak hours/Replication period
Bandwidth ≥ Peak-hour bandwidth (at least 64 Mbit/s)
Mbit/s) (at least 10 Mbit/s)
Active-Active Active-Passive Geo-Redundant
Integrated SAN and NAS A-A Solution

DC A DC B
Solution and Technologies
Host application cluster
(shared volumes • High-reliability service-level A-A architecture ensures zero data loss
mounted to HyperMetro
file systems)
or interruption in case a DC is down (RPO = 0, RTO ≈ 0).
SAN NAS NAS SAN • Concurrent and balanced workloads at both sites ensure user's
WAN access to the closest resources.
Storage network between storage SAN • No gateway, streamlining networking (cost and latency) for 30%
IP&FC systems and hosts IP&FC
higher performance.
Production Cross-DC access Production • Dual arbitration mechanism with quorum server and static priority
storage Real-time data mirroring and storage
dual-write
improves system reliability.
RoCE/FC/IP
• Optimized storage protocol, with fewer cross-site write interactions
(2 to 1) to boost A-A performance by 25%.
• Cross-site RoCE communication supports RDMA to slash latency and
IP network IP network
protocol stack processing overheads, while boosting bandwidth.
• Optimized storage, databases, virtualization, network, and
Active quorum server Standby quorum transmission ensure consistent service experience between local
server
Notes: and remote site (100 km).
1. The frontend of NAS HyperMetro supports only IP networks and not support Fibre • Scalability to geo-redundant 3DC DR solutions without service
Channel networks. interruption.
2. For HyperMetro, RTO is close to 0 and this is because HyperMetro arbitration needs
time if a site is faulty.
Active-Passive DR Solution
Solution and Technologies
⚫ Integrated SAN and NAS active-passive DR solution ensures the
WAN consistency of databases and files.
⚫ Heterogeneous storage resource consolidation in the
production center improves the utilization of the existing
Production center DR center storage arrays while avoiding storage vendor lock-in for DR
system construction.
V V V V V V DR management V V V V V V
⚫ Supports interconnection across high-end, mid-range, and
M M M M M M
M M M M M M entry-level storage systems, and heterogeneous server
Virtualization/ virtualization, reducing investment on devices in the DR center.
Middleware/App ⚫ I/O-based data replication, RPO ≈ 0.
⚫ Automatic or manual switchover between synchronous and
Database asynchronous replications, cutting down the impact of link
jitter on services.
⚫ Supports synchronous and asynchronous replication of
IP&FC VMware vVols 2.0.

IP&FC
⚫ Dedicated DR management software delivers visualized DR
Synchronous/ management and one-click DR drill and recovery.
Asynchronous
replication
Geo-Redundant 3DC DR Solution

Cascaded architecture
Production center Intra-city DR center Remote DR center Solution and Technologies
A
Active-
active/Synchronous/A Asynchronous SAN
SAN SAN replication ⚫ Resilient design with 3 copies and larger DR scope.
synchronous
A replication A' A'' ⚫ Smooth expansion from two to three (cascaded) data
centers, with flexible parallel deployment and multi-
Parallel architecture copy DR.
Production center Intra-city DR center
⚫ Supports the 3DC DR star, ensuring uninterrupted data
A Active-
active/Synchronous/As
replication in the event of a DC failure.
SAN SAN
ynchronous replication Remote DR center
A A' ⚫ Replication interoperability among high-end, mid-
Asynchronous
replication
range, and entry-level storage arrays, with mix-and-
SAN
match selection for DR center devices to increase ROI.
A"
⚫ Visualized management of DR services and topologies,
DR star including one-click DR drill and failover to simplify the
Production center Intra-city DR center
management and maintenance.
A
Active-
SAN
active/Synchronous Standby
SAN Remote DR center
replication
A A'
Asynchronous
replication
SAN
A''
intelligent world.


portfolio, new technology, etc. There are a number of factors that could
cause actual results and developments to differ materially from those
expressed or implied in the predictive statements. Therefore, such
information is provided for reference purpose only and constitutes
neither an offer nor an acceptance. Huawei may change the information
at any time without notice.
Huawei OceanStor Dorado All-Flash Storage
Pre-Sales Training
Contents
1. Introduction to OceanStor Dorado
2. Highlights of OceanStor Dorado

• Ever Fast
• Ever Solid
• Intelligent
OceanStor Dorado Positioning
Overall OceanStor Dorado all-flash unified storage sets a new benchmark for the industry with the highest stability, best-in-class SAN and NAS performance, and intelligent, efficient
Positioning management and O&M.
• The OceanStor Dorado 2000/3000 is entry-level storage products. Its cost-effectiveness is its competitive edge.
Product • The OceanStor Dorado 5000/6000 is positioned as a mid-range storage system, outperforming competitors with its high-end advantages and high-density form factors.
• The OceanStor Dorado 8000/18000 is positioned as a high-end storage system. Its stable architecture, #1 performance, and intelligence are key to expanding high-end
Positioning markets and setting benchmark projects of core NAs.
NAS
• It is positioned as a high-performance NAS product and provides abundant enterprise-class features to meet customers' demanding file storage requirements.
Positioning
High-end competitiveness
Unified SAN and NAS storage Rock-solid architecture + leading intelligent tech
High-end advantages + high-
density form factors
Recommended
Two levels for attack and Best-in-Class
Configurations defense
OceanStor Dorado 8000 OceanStor Dorado 18000
Entry-level all flash
OceanStor Dorado 5000 OceanStor Dorado 6000 Building competitiveness based on high-end architecture
Breaking into core applications based on NVMe/intelligent tech
OceanStor Dorado OceanStor Dorado
2000 3000
OceanStor Dorado Product Overview
Entry-Level Mid-Range 高端 High End
OceanStor
Dorado 18000
OceanStor
OceanStor OceanStor OceanStor Dorado 6000
Dorado 2000 Dorado 3000 Dorado 5000
Entry-Level Mid-Range High End

OceanStor Dorado
Type OceanStor Dorado 3000 OceanStor Dorado 5000 OceanStor Dorado 6000 OceanStor Dorado 8000 OceanStor Dorado 18000
2000
Height / Controllers of
2U/2C 2U/2C 2U/2C 2U/2C 4U/4C 4U/4C
each Engine
Controller Expansion 2-8 2-16 2-32 2-32 2-32 2-32
Maximum Disks 400 1200 3200 4800 6400 6400
Cache/Dual Controller 128GB 128G/192G 256G/512G 1024G 512G/1024G/2048G 512G/1024G/2048G
8/16/32G FC,
8/16/32G FC, 8/16/32G FC, 10/25/40/100Gb Ethernet，25/100Gb
Front-end ports 10/25/40/100Gb Ethernet， 8/16/32G FC, 10/25/40/100Gb Ethernet，25Gb RoCE
10/25Gb Ethernet， RoCE
25Gb RoCE
Back-end ports SAS3.0 SAS 3.0/100Gb RDMA
Lead the Market with Cutting-Edge Huawei All-Flash Storage
Gartner Magic Quadrant:

Market share in China Global market share
Leader
others 23.8%
9.7%
2022 Huawei
Global ranking: No.2
Toyou
6.4%
48.6%
Lenovo
6.7% 19.2%
2016: Leader
2015: Top in Challenger 10.3%
9.1%
DELL EMC
7.9% 7.4%
8.2%
2014: Challenger
Inspur
2010: Niche Player 9.9% Dell EMC Huawei HPE NetApp Pure Storage IBM
H3C
10.6%
Source: Gartner Magic Quadrant for Primary Source: IDC AFA Storage System Market Overview Source: Gartner Market Share Report, 2022 Q4
Storage (China), 2022 Q4
Huawei All-Flash Storage Gains Industry Renown
Interop Grand Prize 2021 Gartner Peer Insights Ranked Recommended by DCIG
The Server & Storage category Customers' Choice for Primary Storage Arrays Listed in Top 5 High-end Storage Array List
A leading technology research and A global IT consulting organization

A major event in global business consulting company and provides the known for extensive analysis and
technology that recognizes the world's most trusted source of objective reviews in enterprise storage
most competitive products independent IT research and advice. and cloud computing
ESG: High Availability, Optimal Performance, Extremely Low Latency, and
78% 5-Year TCO Savings with OceanStor Dorado
Enterprise Strategy Group is an IT analyst, research, validation, One of the lowest response times while maintaining high performance
and strategy firm that provides market intelligence and actionable
insight to the global IT community.
81 µs response time
Three out of four controllers offline in the test — I/Os run normal, and
response time is < 20 µ sec
Controller B Offline
Controller A Offline 220,000 IOPS
Controller C Offline
18 µ sec 20 µ sec
Up to 78% lower TCO over 5 years
Controller C Online
Controller A Online
Controller B Online
Evaluator: Customers evaluating NAS storage systems should consider
Huawei OceanStor Dorado systems
An IT analyst firm dedicated to providing independent expertise on Data and Information Management.
Deliver high I/O rates with low latency for EDA and OLTP, and provide high It balances loads among controllers in a single file system, so no file system
throughput for the medical image workload. belongs to any controller.
OceanStor Dorado V6 NAS

Workload Performance Comparison
Throughput in MB/S
24,618
21,484
10,029
EDA: 16K I/O OLTP-DB: 16K I/O MEDICAL: 1 MB I/O

-100% FILE READ -70-30 R/W -100% WRITE
Overall, OceanStor Dorado NAS systems were found to provide comprehensive set of NAS features along with excellent performance and reliability required for enterprise
application workloads. Customers evaluating NAS storage systems should consider Huawei OceanStor Dorado systems.
OceanStor Dorado All-Flash Storage Highlights
Ever Fast Ever Solid Intelligent&Efficient
Industry-leading SmartMatrix fully Industry-leading SAN&NAS solution

performance and interconnected architecture Intelligent full-lifecycle management
latency for always-on applications Full-lifecycle ransomware protection
21M IOPS and 0.05 ms latency Tolerates failure of 7 out of 8 controllers Intelligent O&M
30% higher NAS performance than Provides active-active solution for SAN Industry-top ransomware detection
industry benchmark and NAS capability
Contents

• Ever Fast
• Ever Solid
• Intelligent
Ever Fast
11
Huawei Confidential
Innovative E2E Acceleration Platform Achieves 21M IOPS and 30% Higher
NAS Performance
Efficient protocols
E2E NVMe for a high-performance
expressway NVMe
Innovative E2E acceleration for
leading performance
Distributed file system
Distributed file system maximizes system
performance Distributed file systems
Accelerated algorithms
SSD-controller synergy with intelligent algorithms
maximizes all-flash performance FlashLink®
Uncompromising Performance with Innovative FlashLink® Algorithms
®
FLASHLINK intelligent algorithms
Controller
Core Reconstruction 81.79%
53.13%
001
0 Core
Many-core Core
0 Ordered 1 2 3 4 5
0 0 scheduling 300 min/TB 15 min/TB
1
1 010 1 by reboot
0 101 Core sequence
0 1
1
1 0010
0
0 0
1
0 0 1101 0
Smart00SSD
0
00 enclosure
10
1
1 1 010 1
Many-core algorithm Service grouping Intelligent cache algorithm
11 11
001 0101 0 2x computing power 20x reconstruction speed 50% cache hit ratio
011 10 11 0 0
101 1010 1
1 0 0 01
1 1
011 01101 0
1
1 1010 1
1 1
0 0101 0
011
Data read
1 11 0
0
1 Metadata
Data write
1 0
0SSD00
New data
11 11 Garbage Advanced feature
1 1 collection data Reconstruction

Garbage collection
Full-stripe writes Multi-streaming data separation Global I/O priority adjustment

Less write amplification Less garbage collection Lower latency
Smart SSD Enclosure Implements Service Splitting and Computing Power
Unlocking
Smart SSD enclosure
Controller enclosure
Front-end Front-end Front-end Front-end 30% computing power sharing
interface module interface module interface module interface module
• Each smart SSD enclosure has two control boards with built-in
Data Data
reconstruction Controller
Controller ControllerreconstructionController computing platform and memory resources
• Smart SSD enclosures take over some workloads like data

Back-end Back-end Back-end Back-end
interface module interface module interface module interface module reconstruction from array controllers
Impact on controller performance

Data reconstruction time
(maximum IOPS)
30 min/TB 15 min/TB 15% 5%

Smart SSD enclosure + computing power Smart SSD enclosure + computing power
platform platform
Data reconstruction bandwidth Controller CPU utilization
80 MB/s 200 MB/s 70%
LUN Active-Active: Fully End-to-End Symmetric Architecture
Active-Active Architecture
Host • All products support host access in active-active mode.
Requests can be evenly distributed to each front-end link.
• All products support LUNs without owning controllers, which
SAN is easy to use and load balancing. (The LUN is divided into
several slices, and the slices are evenly distributed to all
controllers using the DHT algorithm.)
• High-end products provide front-end intelligent shared I/O
modules to directly send host I/O slices based on LUNs to
target controllers, reducing latency.
Hash sharing
DHT
Global Cache
• the LUN writes the I/O directly to the cache of the
controller where the shards reside and returns a response
to the host.
Global Cache
• Intelligent read cache for all controllers participates in
data and metadata prefetch and cache hits for all LUNs
Global Storage Pool

• Storage pools are evenly distributed on all controllers.
… … RAID 2.0+ is used to manage all SSDs in the storage
pool.
Distributed File System with Global Sharing Enables Fast NAS Performance
DIR
Distributed
data balancing File File File
Intelligent balancing vNode
Directory balancing algorithm: Directory and file co-
Faster file systems with all controllers ownership processing:
Zero cross-core and -controller
System architecture overheads
Readdir 8 KB
Filelist Filelist Filelist Metadata
/A /B /C /D Global Cache
Intelligent cache Intelligent metadata
prefetch/elimination Metadata compaction algorithm:
Distributed file system algorithm: Accelerates index Greatly improves cache and
file search prefetch efficiency
Controller 1 Controller 2 Controller 3 Controller 4
Global resource pool
Intelligent layout Metadata sequence table Metadata hash table layout:

Accelerateslayout:
file traversing Accelerates file locating
(Readdir) (Lookup)
Provides Directory Balancing and Performance Modes to Fit into Different
Scenarios
Host Host
NFS/CIFS NFS/CIFS
Client Client
Create /A, /B, /C, Create /A, /B, /C,

and /D directories. and /D directories.
Distributed file system Distributed file system

NFS/CIFS NFS/CIFS
/A /B /C /D A B C D
/A/A1 /A/A2 /A/A3 /A/A4 /A/E /B/G /C/I /D/K
/A/A1/A11 /A/A2/A22 /A/A3/A33 /A/A4/A44 /A/F /B/H /C/J /D/M
CPU CPU CPU CPU CPU CPU CPU CPU
• The directory balancing mode eliminates • The performance mode ensures no cross-core
bottlenecks from a single controller or CPU. overhead for a single directory.
Online transactions: 5x more TPS
OceanStor
5x Better User Experience
Dorado 6000 57,000 TPS 40% higher in large files 30% higher in small
Vendor E AFA and small blocks files and small blocks
11,500 TPS
700
Test conditions: dual-controller, 40 x 3.84 TB SSDs,
SwingBench OE2 transaction simulation system
500
*TPS: transactions per second
380
Report queries: 33% shorter batch processing
300
180
120
5.6 hr 3.8 hr Unit: K OPS Unit: K OPS
Test conditions: dual-controller, 512 x 20 GB Test conditions: dual-controller, 40 million x 8 KB files,

files, 8 KB block size, 7:3 mixed read/write 5-level directories, 8 KB block size, 7:3 mixed read/write
Vendor E AFA OceanStor
Dorado 6000
Test conditions: dual-controller, 40 x 3.84 TB Equivalent with industry
50% higher in bandwidth-
SSDs, report query simulation system, 3 TB data benchmark in operational
intensive scenarios
I/O scenarios
VDIs: 80% faster application response 300
5000 5000
5s 5.1s 200
4.7s
Vendor E AFA 2800
1.8s OceanStor Dorado 6000
1.5s
0.9s 80
Unit: 100 MB/s Unit: OPS
Test conditions: dual-controller, 100 x 3.84 TB Test conditions: dual-controller, 512 x 20 Test conditions: dual-controller, and
GB files, 1 MB I/O sequential read mkdir/rmdir/create/delete/readdirplus
SSDs, 8 TB per LUN, 50 GB per VDI
Contents

• Ever Fast
• Ever Solid
• Intelligent
12,000+ DCs
13-year stable operation on the live network
Ever Solid Applications with 5 Reliability Layers
Cloud ▪ Effortless data backup
▪
99.9999%
Active-active solution for SAN and NAS ensures always-
on services
Solution ▪ FlashEver without data migration
system-level reliability ▪ Comprehensive enterprise-class features

▪ Tolerance for simultaneous failures of 3 disks
System ▪ Reconstruction of 1 TB data in 15 minutes
▪ Enclosure-level redundancy tolerates the enclosure failure
▪
99.99999%
Dynamic Reconstruction Ensures Redundancy Ratio
▪ Tolerance for failures of 7 out of 8 controllers with

Architecture SmartMatrix fully-interconnected architecture
solution-level reliability ▪ E2E active-active design
▪ Global wear leveling

SSD ▪ Huawei-patented global anti-wear leveling
Architecture Reliability-SmartMatrix Sets a New Benchmark
BIM**
Global cache
CPU CPU CPU CPU

Ever solid architecture
FIM*
• Tolerates failures of up • Tolerates failure of 1

to 7 controllers controller enclosure
CPU
CPU
Global cache
SmartMatrix OceanStor
Global cache
CPU
Vendor E Vendor H
CPU
BIM
FIM
BIM
FIM
Tolerance Dorado Interconnection: Interconnection: front-
CPU
Full interconnection:
CPU
back-end (2- end (full); back-end (4-
front and back-ends controller) controller)
CPU
CPU
1-controller failure
2-controller
failure
FIM
CPU CPU CPU CPU

7-controller failure
Global cache
BIM
*Front-end interconnect I/O module (FIM)
**Back-end interconnect I/O module (BIM)
Architecture Reliability-Continuous Front-End Link in the Event of Controller
Failure Servers
Shared Shared Shared Shared
frontend frontend frontend frontend
Data
Controller Controller Controller Controller
Shared Shared Shared Shared FC or ETH

backend backend backend backend
network
• The FIMs are linked to servers independently regardless of FIM FIM FIM FIM
controller failures.
Data
• Each I/O goes to a specific storage controller through the

backplane. Back plane
Controller Controller Controller Controller

• Faulty controller I/Os redirect to another functioning controller
for a continuous link between the FIM and the server.
Architecture Reliability-Non-Disruptive Upgrade in Seconds with Higher
Stability and Simplicity
Host
✓ Modular software architecture
◼ 94% of components are in user mode and can be upgraded within 1s.
FC switch ◼ Components on the I/O path can be upgraded within 1s.
✓ Continuous link and host
unaware of upgrade ✓ FIM:
◼ Connects 4 internal links to 4 controllers in an enclosure
◼ Provides 1 communication link to the host through the front-end port
✓ I/O components ◼ Switches the service link to another controller when any controller restarts during
upgraded within 1s an upgrade. The host is unaware of the upgrade, and the communication link is
uninterrupted.
✓ The firmware supports loose mappings and patch upgrades. In extreme scenarios, the
FIMs can ensure host linkup.
Competitor E Competitor H Huawei

I/O process I/O process I/O process I/O process
System mgmt. Device mgmt. Configuration

NMS process
Time (s) 3–5 > 10 1
process process mgmt. process Component-
based √
Continuous host
User mode upgrade
link
Yes Yes Yes
Time (minute) 132 90 30

Kernel mode Rolling
upgrade Continuous host
No Yes Yes
link
Ctrl 0 Ctrl 1 Ctrl 2 Ctrl 3
Restarted during the upgrade
System Reliability-HyperZoom Dynamic Reconstruction Ensures Redundancy
Ratio
RAID 6 (7+2) RAID 6 (6+2)
A0 B0 C0 D0 E0 F0 G0 P0 Q0 A1 B1 C1 D1 E1 F1 P1 Q1
• If the disk number is less than the number of member disks required in a RAID group, the system automatically reduces the number of member disks in a RAID group.
• When a disk fault is rectified and the total number of available disks in the disk domain meet the RAID requirement, the number of RAID data columns automatically
restores to the normal value.
System Reliability-HyperCDP Has Industry-Leading Interval of 3 Seconds
Interval: 3s for LUN and 15s for FS
LUN/FS TP TP+1 TP+2 … TP+N
CDP snapshots CDP 0 CDP 1 CDP 2 … CDP N
Snapshot copy creation Max. 60,000 for LUN

Max. 4,096 for FS
Rollback
Snapshot copy Snapshot copy
Periodic policies with shortest interval More intensive and persistent data protection
✓ Supports periodic policies by day, week, month, or specific interval ✓ HyperCDP only saves the corresponding time points, surpassing
for customized backup. common snapshots with stronger and longer protection.
✓ The multi-time-point and ROW technologies create a data copy (at ✓ A single LUN supports up to 60,000 HyperCDP snapshots. The
a certain point in time) of a source LUN at a minimum interval of 3s minimum snapshot interval is 3 seconds, providing continuous
and that of an FS at 15s. protection for three days.
✓ Each CDP snapshot matches a time point of the source LUN/FS.
Solution Reliability-Active-Active Solution for SAN and NAS Meets the Most
Demanding Reliability Requirements
ERP Lightning-fast and rock-solid

CRM BI
File Sharing • Gateway-free: SAN and NAS convergence, fewer nodes,
and simplified management
• SAN active-active: Load balancing between sites, RPO = 0,
RTO ≈ 0, end-to-end solution, 50% higher all-IP A-A
HyperMetro performance compared to traditional IP solution
Gateway-free active-active • Industry's only NAS A-A solution
Production center A Production center B
HyperReplication HyperReplication
Easy-to-scale
OceanStor storage
OceanStor Dorado system • Scalability to 3DC (SAN) improves reliability.
• Serial, parallel, and star networking (SAN) meets the most
demanding requirements for enterprise reliability.
DR center
System Reliability: Comprehensive Ransomware Protection Solution*
Intra-city/ Remote • Automatic detection and alarm

Local DC
DC reporting: Ransomware can be
detected in seconds through the
Isolation zone
AI-based ransomware detection
Test host BCManager 5
(Optional) 6 capability, I/O behavior- and
Active-
IP switch IP switch
active/
Replication snapshot comparison-based
detection, and suffix-based
identification.
• High-density lossless secure
Firewall
1 4 5 (Optional)
1 2
snapshot: Secure snapshots are
3 7 5 3 7 automatically created every 15
OceanStor Dorado
seconds (3 seconds for SAN),
ensuring an RPO within seconds.
Email server
Host link Management link Replication link
Key 1 Detection and analysis* 2 File blocking* 3 Secure snapshot 4 Air Gap
technologies 5 Replication link encryption 6 Protocol encryption* 7 Storage encryption * Indicates NAS-specific functions.
Contents

• Ever Fast
• Ever Solid
• Intelligent
28
Huawei Confidential
Native Unified Storage Architecture Provides the Same I/O Path for NAS and
SAN Without Extra Overhead
iSCSI/FC/NVMe-oF NFS/CIFS
Protocol layer
SAN and NAS protocols are isolated. iSCSI, NFS, and CIFS support DTOE offloading.
Data service
Smart Series: Cache, Tier, Quota, Migration, Virtualization Data service layer
Hyper Series: Replication, Vault, Metro Value-added features and data protection features based on LUNs and file systems
Global cache
Global cache SAN/NAS global cache sharing
Cache for data, metadata, and deduplication fingerprints
Space management layer

Block File
Active-active data space management for LUNs
Space management for NAS distributed file systems
Hot and cold data Global wear leveling ROW full-stripe write
separation Storage pool
Global Post-process Global garbage SAN and NAS share the FlashLink-based global storage pool, which provides global
deduplication similarity-based collection
deduplication deduplication, compression, similarity-based data reduction, global wear leveling, and global
garbage collection.
Storage pool
Cloud Native Container Storage Base-Quick Deployment and Powerful
Functions
Container Management Platform (Provided by Partners)
Containerized application layer Advanced technology: Enterprise-class container management platform.
Professional support: First-class technical support for complex issues.
Professional CDR Container DR Management
Container Reduced backup cost: Application-level backup and recovery without additional backup
management software, reducing TCO.
platform Fast recovery: Services are available immediately after recovery, ensuring data security.
Huawei CSI* Huawei CDR*
Standard CSI Support for Efficiency and Simplicity
Easy to use: Standard CSI, on-demand storage resource provisioning, easy to use.
Computing Network Fast service provisioning: Batch service provisioning is 30% faster than the industry
(ECS/BMS/ARM/GPU)
Storage (VPC/EIP) average.
Fast and Stable Data Storage for Zero Loss of Container Data
High performance: NAS performance is 30% higher than the industry average.
High reliability: Multiple reliability designs and automatic failover ensure service
OceanStor Dorado continuity.
Provided by High security: Industry-leading SAN/NAS ransomware protection, protecting container
Huawei *CSI: container storage interface data from ransomware threats.
*CDR: container disaster recovery
Flash Storage Convergence for Different Types, Levels, and Generations
⚫ Different architecture
OceanStor Dorado
⚫ Different generations
2000/3000/5000/6000 ⚫ Low data mobility
HyperReplication
OceanStor Hybrid Interconnected services for free data mobility
OceanStor Dorado Unified management for 50% lower complexity
Flash Storage 8000/18000 High-end + entry-level products for customer savings
Different types: OceanStor Dorado and OceanStor

hybrid flash
Different levels: OceanStor Dorado High-end, mid-
range, and entry-level
OceanStor Dorado V3 Different generations: OceanStor Dorado V3
SmartMobility for Free Data Flow, Reducing TCO Throughout the Lifecycle
Hybrid flash storage
NAS client
Warm and cold Tiering policy
File system data: 300 TB
File name
Option 1
File size
Read/Write Object storage
Creation time
Option 2 Modification time
Warm and cold
Total: 500 TB data: 300 TB
Access time
Hot: 200 TB
OceanStor Dorado Contents
All-flash storage Cloud
Option 3 File type
Warm and cold

OBS data: 300 TB
➢ Experience: The client only needs to access the all-flash storage and does not need to consider the file storage location.
➢ Direction: Cold data is automatically migrated to warm and cold data storage media. When a file is accessed, the file is automatically migrated to the all-flash storage.
➢ Target: Files can flow to NAS storage, object storage, or public clouds.
Three Intelligent Layers: Full-Lifecycle Data Management for the Intelligent
Era
Intelligent
Cloud Intelligent Expert experience
prediction
DME IQ
Model training Call Home service Cloud Brain for intelligent O&M
Intelligent model training
Automation engine Prediction engine DME*

Center Intelligent
Execution engine and automatic
Intelligent model application Policy engine Analytics engine
management
Device Intelligent DeviceManager

Self-learning Self-optimization
Intelligent Intelligent devices and simplified
hardware and algorithms configuration
* DME: Data Management Engine
DeviceManager: Easy-to-Use, Intelligent Device Management
Integrated GUI
Simplified GUI
Unified portal for multiple services: >50%
Block File Name Object* higher management efficiency
Step 1 Step 2 Step 3 Simplified deployment

Create LUNs Create hosts Create mapping views I/Os provisioning in three steps:
4 clicks 1 click 6 clicks 2 clicks 1 click

64% faster deployment
Create protection groups

Simplified protection
Protection group management from the app perspective:
Snapshot Clone consistency Replication Active-active

75% fewer operations
consistency group group consistency group consistency group
DME-Intelligent Data Lifecycle Management Platform
IT management platform
Automation Monitoring Report

platform platform platform Multi-cloud connection
Data center status clear at a glance 6 types
Visual • Dashboard and report Risk identification in
• Data center online insight advance
DME Storage
Visual Automatic Intelligent
Shorter solution design Improved proactive O&M
Precise investment planning
period efficiency
Accelerating service change

Enhanced resource
utilization
Improved service rollout
efficiency
Shorter troubleshooting
period
• Service change simulation
10x
Automatic
• Automatic service provisioning Improved resource
change efficiency
Converged resource pool • Data security and protection
Stable service running

• One-stop intelligent management
… • Global intelligent search 90%

Service host Switch Virtualization Intelligent
• Intelligent health check Shorter fault
locating period
• AI-based trend prediction
OceanStor OceanStor Heterogeneous
Dorado Pacific
OceanProtect • Intelligent assistant fault diagnosis
storage
DME IQ-Cloud-based Intelligent O&M Platform for Customer Data Centers
DME IQ uses AI technologies such as machine learning and deep learning to perform big data analysis. It proactively helps customers monitor, analyze, and plan IT infrastructure such as Huawei
storage devices and servers. It implements proactive and predictive O&M anytime and anywhere, reducing O&M costs and improving O&M efficiency. Therefore, it provides one-stop O&M
solutions for customers.
Planning Centralized Intelligent Intelligent 24/7 Online, Improving O&M Efficiency

simulation O&M monitoring prediction
• Web or mobile access, anytime, anywhere

• Installation-free, one-click apply, instant provisioning
Proactive
O&M
Intelligent Prediction, Identifying Risks in
Web
Advance
DME IQ
HTTPS • Disk fault prediction 14 days
in advance • Intelligent analysis of VMware
• Performance bottleneck performance issues
prediction 90 days in • Quick demarcation of Oracle
Mobile HTTPS AI-powered advance performance issues
analysis • Capacity trend prediction 1
year in advance
Device Planning Simulation and Precise Allocation,

Reducing Customers' TCO
• Service rollout simulation
• Service scaling simulation
Service • Service migration simulation
Storage Server Virtualization
planning
intelligent world.


Huawei New-Gen OceanStor Hybrid Flash Storage
Pre-sales Training
Contents
1. Product Positioning and Specification Comparison
2. Software Architecture Upgrade
3. Product Highlights
New-Gen OceanStor Hybrid Flash Storage: Product Positioning
➢ Replaces OceanStor V5 Kunpeng
➢ Covers 6 mid-range or higher models — OceanStor 5310/5510/5610/6810/18510/18810
➢ Tailored for the hybrid flash market (i.e. the role of SSDs and mandatory config.) and all-
OceanStor
flash evolution Dorado 18000
512 GB/1 TB/2 TB
➢ Integrates the advantages of OceanStor Dorado and OceanStor V5 Kunpeng
OceanStor
Dorado 8000
High-end model OceanStor 18510
(entry point) OceanStor 18810
512 GB/1 TB/2 TB
OceanStor 1 TB/2 TB
OceanStor OceanStor
Dorado 3000 Dorado 5000 Dorado 6000
128/192/256 GB 256/512 GB 1024 GB
OceanStor 6810
High-end model
(entry point)
OceanStor 5310 OceanStor 5510 OceanStor 5610
64/128 GB
128/256 GB 384/512 GB 768/1024 GB
OceanStor OceanStor
2200/2220 2600/2620
32 GB 64/128 GB
Key Features: New-Gen OceanStor Hybrid Flash Storage vs. OceanStor V5
Upgraded hardware and architecture innovation
compared to OceanStor V5 Kunpeng
Category OceanStor V5 Kunpeng New-Gen OceanStor Hybrid Flash Storage
LUN ALUA A-A LUN A-A
Architecture
File system: A-P architecture File system: distributed file system
High-end models tolerate 3/4 controller
Architecture Reliability High-end models tolerate 7/8 controller failures
failures
Kernel-mode upgrade. Controller restart User-mode upgrade. No restart required in 95% of
Upgrade
required scenarios
Hardware
Host protocol NoF is not supported. Supports FC-NVMe and NVMe over RoCE (NoF+)
capabilities
COW mode: snapshot performance
deteriorates significantly. ROW-based lossless mode:
Snapshot
High-density snapshots and cascaded Supports high-density and cascaded snapshots.
OceanStor snapshots are not supported.
A-A solution for SAN
5310/5510/5610/6810/18510/18810 Metro
Industry's only A-A solution for SAN and NAS
Function clustering
A-P solution for NAS
Supports RAID 5, RAID 6, RAID-TP, and RAID zoom
Supports RAID 0, 1, 3, 5, and 6. Tolerance for
RAID (dynamic RAID reconstruction). Tolerance for 3 disk
up to 2 disk failures
failures
Built-in
Not supported Supported
container
Contents
Two Architecture Upgrades of the New-Gen OceanStor Hybrid Flash Storage
OceanStor V5 New-Gen OceanStor Hybrid

Flash Storage
Write in Place Redirect on Write

COW snapshot ROW snapshot
ALUA A-A Fully balanced A-A

architecture architecture

ALUA File System
OceanFS
Write-in-Place Architecture Causes Huge Write Penalty, Impacting Hard
Disks with Insufficient Performance
1111
Parity bit
0110 1101 0101 0010
RAID Level Write

Take RAID 5 as an example. If 0110 is changed to 1111 and the Write-in-Place Penalty
architecture is used: 0 1
• Read the original 0110 and perform exclusive OR (XOR) with the new data
1111. 1111 XOR 0110 = 1001 1 2
• Read parity bit 0010 and perform XOR with the above result. 1001 XOR
0010 = 1011 5 4
• Then write 1111 to the original position and write 1011 to the parity bit.
6 6
In conclusion, writing 1111 once will generate two reads and two writes, 10 2
resulting in 4x the write penalty.
New-Gen OceanStor Hybrid Flash Storage Uses ROW-based Large Block
Sequential Writes to Solve HDD Performance Bottlenecks
Data to be
modified
B' D' E' Snapshot
S
Data
written A B C D E P1 P2
Snapshot
S The latest data
N
Data written
The latest data
A B C D E P1 P2
Snapshot
S N
New data
Data after
B' D' E' X1 X2 P1' P2'
modification A B' C D' E' P1' P2'
ROW does not cause write penalty and data migration.
Snapshot ROW + large block sequential writes convert random I/Os to sequential large I/Os,
protection zone Write penalty generated by which is more suitable for HDDs.
B D E writing data in place; High-density snapshots can be created without migrating original data.
High data migration costs; ROW increases the amount of metadata and requires frequent rewriting of
Snapshots greatly affect metadata.
Write-in-place performance and cannot be used
as high-density snapshots. ROW + large block sequential writes
OceanStor V5 SAN
New-Gen OceanStor Hybrid Flash Storage
New-Gen OceanStor Hybrid Flash Storage Uses the Symmetric A-A
Architecture for True Load Balancing
LUN/FS ownership: I/O requests
No LUN/FS ownership: I/O requests
from hosts to LUN-A/FS-A can be Host
delivered only through
from hosts to LUN-A/FS-A are Host
distributed to multiple controllers for
controller A.
load balancing.
Forwarding
Controller A Controller B Controller C Controller D Controller A Controller B Controller C Controller D

Forwarding
Forwarding
LUN-A/FS-A
LUN-1/FS-1
ALUA A-A architecture Symmetric A-A architecture
OceanStor V5 New-Gen OceanStor Hybrid Flash Storage
Architecture Upgrade-100% Performance Improvement
Performance comparison in
SmartAcceleration OceanStor V5
typical configurations
48 NL-SAS HDDs + 4 SSDs
New-Gen OceanStor
Hybrid Flash Storage
ROW-based Large Block 180%
110%
Sequential Writes
Global data write optimization
Flexible Convergence of 230%

Heterogeneous Media
Cache Tier Optimal cost-performance balance
Global Cold/Hot Data

Perception
Multidimensional feature 8 KB I/O 64 KB file 2 GB file and 8 KB I/O
Reads/writes: 7:3 Reads/writes: 6:4 Reads/writes: 7:3
learning of neural networks *Test results from Huawei Lab
Architecture Upgrade-Extensive Value-Added Features (VAFs) and Minimal
Impact on Performance After VAFs Are Enabled
Advanced data protection functions Minimal impact on performance (VAFs
enabled)
Performance decrease: 20% to 30% for competitor vs. <10% for Huawei
HyperCDP Snapshot Cascading

High-density snapshot consistency groups snapshot
3s protection interval
OceanStor V5
New-Gen OceanStor Hybrid

Initial performance GC RAID 6 Snapshot
Flash Storage
Competitor Huawei OceanStor
hybrid flash hybrid flash
Test equipment: Dual-controller, 1 TB cache, 48 NL-SAS HDDs + 4 SSDs
Architecture Upgrade-Industry's Only Active-Active Solution for NAS
EDA
PACS CAD
File Sharing
Cross-site directory access balancing
Industry's only
New-Gen OceanStor New-Gen OceanStor
A-A solution for NAS Hybrid Flash Storage Hybrid Flash Storage
HyperMetro
Gateway-free A-A solution
Production center A Production center B
HyperReplication HyperReplication
• RTO = RPO = 0 OceanStor storage system
• Load balancing across 2 data centers

• Smooth upgrade to 3DC/4DC solutions
DR and backup center
Content

Future-Oriented Design that Supports Multi-Workloads
Continuous Evolution, Oriented to Multi-

Workloads with high security
Supports multi-workloads, such as blocks, files, virtualization, and containers.
Cloud backup and private cloud interconnection to support cloud-based evolution.
Multiple security protection methods to protect enterprise value data.
Full Upgrade for Flash-Like Experience

SmartAcceleration in all scenarios and convergence of cache and tier.
SmartMatrix A-A fully load balancing architecture maximizes the performance.
OceanStor
5310/5510/5610/6810/18510/18810 Reduced TCO
No SAS HDD required reduces construction costs; supports reuse of
cross-gen devices.
Intelligent and automatic O&M greatly reduces O&M costs.
Diverse Workloads Flash-Like Optimal TCO
Multi-Workload Support to Meet Diversified Customer Requirements
Data Center
Multiple Application Ecosystems
Traditional core Critical + emerging
services services Supports block, file, container, VM, and
distributed database apps.
DB2 Oracle MySQL MongoDB AIOT

Diverse Cloud-based Evolution
Virtualization Hybrid cloud Container
Supports cloud backup, private cloud, and public cloud-like
BMS VMware
x86 FusionCompute
deployment for diversified cloud-based evolution.
ARM KVM
Multi-level Protection for Regulatory

SAN NAS Compliance
Zero-trust-based permission management, antivirus, WORM,
HDD -> SSD
and log audit functions.
New-Gen OceanStor Hybrid Flash
Storage
All-Scenario Interconnection with VMware Facilitates Fine-Grained

Management
VMware ESXi vCenter
Full COPY
HW-Assisted Locking
VASA Provider DME plugin
Block Zero
Thin Provisioning Unmap
VAAI integration improves the Storage integration supports

VMware vSphere efficiency of data storage VASA integration enables vCenter storage resource provisioning on
operations. to sense storage information. vCenter
VM VM VMware Metro clustering

SRM
eSDK Storage SRA
PE PE Disaster recovery
plugins
Vvols 2.0 for SRM integration for automatic vMSCs metro clustering for
VM-level resource remote and unified management of 99.9999% virtualization
replication virtual DR processes reliability
VMware ESXi
VMware ESXi
FullCOPY
FC/iSCSI HW-Assisted Locking
LazyClone/
FC-NVMe Block Zero
FullClone/
NVMe over RoCE Thin Provisioning Unmap
Reserve Space/ UltraPath for VMware
Extended Statistics
New-Gen OceanStor Hybrid VAAI-SAN supports four

networking modes to PSA integration for path
Flash Storage
improve efficiency of data management and load balancing in
VAAI-NAS the ESX OS
storage operations.
Storage for Cloud-Native Container-Powerful Functions and Quick Deployment
Internet Enterprise Finance Media Education
Application auto
Control plane HA
scaling
Kunpeng container Storage plug-ins Image scanning BMS container
Huawei CSI feature highlights:
Automated Volcano batch Automated Permissions
Node auto scaling GUI/CLI/API
configuration processing deployment management
✓ Standard interface definition for interconnection with multiple
container platforms
Enhanced features of containerized applications
✓ The out-of-tree model is decoupled from K8s, and supports
containerized deployment, which is more secure and flexible.
✓ Powerful functions include volume and snapshot management,
and support continuous upgrade.
Open-source native container platform

Huawei CDR feature highlights:
gRPC
Storage ✓ Reduced backup cost: Application-level backup and recovery

Compute (ECS/BMS/ARM/GPU) Networking (VPC/EIP)
(Huawei CSI, Huawei CDR) without additional backup software, reducing TCO.
✓ Fast recovery: Services are available immediately after recovery,
ensuring data security.
New-Gen OceanStor Hybrid Flash Storage
CloudBackup-Data Backup and Smooth Data Migration to the Cloud
HUAWEI CLOUD AWS S3 • With CloudBackup, data can be backed up to public cloud object storage
OBS
(currently, HUAWEI CLOUD OBS and AWS S3 supported) or backup
Incremental Difference
storage in a local data center.
backup of comparison
differential data
History Current • Based on SmartContainer deployed on storage, CloudBackup can directly
snapshot snapshot
access storage file systems and read-only snapshots, reducing the cost of
CloudBackup
purchasing backup servers.
SmartContainer
• Configurable full backup and incremental backup policies
Read-only
FileSystem snapshot
• Users can import cloud copies and recreate copy database information
for DR, service migration, and test & dev.
Anti-Ransomware Solution: Flexible Combos of 7 Protection Measures to

Prevent Ransomware Attacks
Network layer Storage layer Backup layer
NAS virus scanning Air gap backup

Periodic antivirus scanning to Backup devices are isolated from
remove risks the network to defense against
virus attacks.
Ransomware attack process:

1. Intrudes user networks.
2. Searches for important files of Encrypted storage Ransomware detection in
Encrypted transmission
users. Encrypted transmission for Encryption before storing copies
replication/Active-active design; prevents information leakage. Identifies infected copies,
3. Delivers targeted ransomware.
no disclosure of sensitive followed by early isolation and
information email warning.
Secure snapshot Secure copy-based recovery

Read-only file system Uses secure copies for rapid
snapshots prevent tampering recovery if primary storage is
and can be used for recovery infected.
at any time.
Storage Array Encryption: Lossless Performance, Multi-Level Data Encryption,

and Full Functional Compliance
Storage Array Encryption
✓ 0 service reconstruction: The operation of storage-level
Application server encryption does not affect applications, so application
reconstruction is not required.
Key management server ✓ Full functional compliance: Supports the AES-256

algorithm and CPU encryption engine.
Storage controller
Data encryption Data encryption engine
✓ Low performance loss: Dedicated encryption
Key lifecycle management.
acceleration based on Kunpeng processors or SSD
Storage disk control chips, achieving zero or low loss of performance
Secure Snapshot: Protecting Backups of Critical Data from Accidental or

Intentional Deletion
Cycle frequency of 3 seconds (LUN)/15 seconds (FS)
LUN/FS TP TP+1 TP+2 … TP+N
CDP snapshots CDP 0 CDP 1 CDP 2 … CDP N
Creating secure snapshots Max = 60,000 for LUN

Max = 4,096 for FS
Rollback
Secure snapshot: Rapid recovery:

◆ ◆
Supports creation of a single secure snapshot and a Infection-free snapshots are available in the
secure snapshot consistency group. production center for recovery In the event of a
◆
Supports conversion from HyperCDP to secure disaster
◆
snapshots; and HyperCDP consistency groups to secure The recovery granularity is as low as 3s (SAN)/15s
snapshot consistency groups. (NAS).
A New Industry Standard for Hybrid Flash Storage: Innovative Hardware,

Advanced Architecture, and Intelligent Algorithm
Advanced architecture
Symmetric active-active architecture

Symmetric A-A architecture leads load balancing between controllers,
improves both performance and reliability.

Distributed file
system The file system is distributed by directory, maximizing storage
computing power.
Intelligent algorithm 100% better performance than previous

generation
SmartAcceleration
Industry's only that allows 7 out of 8
The innovative cold and hot data migration algorithm creates controllers to fail
flash-like experience.
Industry's only A-A design for NAS
Symmetric A-A Architecture for LUN Maximizes Controller Performance
Symmetric A-A architecture

FC/NoF/iSCSI ⚫ Global cache is used, and all controllers participate in
service processing. Workloads of a single LUN are
Global cache balanced among multiple controllers.
⚫ A single LUN can use 90% of controller performance,
and the performance error between controllers does
not exceed 5%.
⚫ A high-speed RDMA network interconnects 32
controllers, scaling controller performance
horizontally.
Global Distributed File System Enables Industry-Leading NAS Performance
DIR
Distributed data
balancing File File File
Intelligent balancing vNode

Directory balancing algorithm: Directory and file co-
Faster file systems with all controllers ownership processing:
Zero cross-core and -controller overheads
Global sharing architecture
Readdir 8 KB
Metadata
Filelist Filelist Filelist
/A /B /C /D Global Cache
Intelligent cache
Intelligent metadata Metadata compaction algorithm:
Distributed file system prefetch/elimination algorithm: Greatly improves cache and prefetch
Accelerates index file search efficiency
Controller 1 Controller 2 Controller 3 Controller 4
Global resource pool
Intelligent layout Metadata sequence table layout: Metadata hash table layout:
Accelerates file traversing Accelerates file locating
(Readdir) (Lookup)
SmartAcceleration: SSD Cache + Tier Convergence for Read Acceleration

in All Scenarios
Database/virtualization applications Database/document Bill imaging/mailbox applications
applications
Cache Cache Cache
SSD cache SSD cache

SSD tier SSD tier
HDD HDD
HDD
Hot data changes frequently. Cold data is read immediately after Data changes from hot to cold based on
being modified. the write time.
SmartAcceleration: Innovative Data Layout Algorithm Accelerates All Hot

Data and Delivers Flash-Like Experience
ROW-based large block

SmartAcceleration sequential writes
Global data write optimization
8 data acceleration scenarios 100% better performance
Reading hot data from fixed addresses
Reading hot data from unfixed addresses
Writing hot data to fixed addresses Flexible convergence of

Writing hot data to unfixed addresses heterogeneous media
Reading hot data after write
SmartCache Cache Tier Optimal cost-performance
Overwriting non-hot data balance
Reading hot data with access frequency
SmartTier (Write in Place)
decaying over time
Writing hot data with access frequency
decaying over time
SmartTier (ROW) Global cold/hot data
perception
Multidimensional feature
learning of neural networks
SmartAcceleration: Elastic Tier/Cache Convergence for Optimal Cost-

Performance Balance
Traditional hybrid flash storage New-Gen OceanStor Hybrid Flash Storage
Tier/cache space isolated Elastic tier/cache converged
SSD cache Performance layer

SSD cache/tier
Data migration based on
access frequency
Periodically
Disadvantages migrated
• The SSD performance downward
layer is exclusively used
SSD tier SSD tier and cannot be shared NL-SAS tier NL-SAS tier
across pools.
• Caches and tiers are
independent, and
SAS tier SAS tier physical space usage is Advantages
inefficient. • Caches and tiers converge into a performance layer,
• Data needs to be enabling efficient cross-pool sharing.
frequently migrated • The cache-to-tier capacity ratio is automatically
NL-SAS tier NL-SAS tier based on access configured by the system, achieving optimal efficiency.
frequency, affecting • Data does not need to be migrated across caches or
Pool 1 Pool 2 performance. tiers, reducing the impact on performance.
SmartAcceleration: Adaptive Access Frequency Perception and Prediction

Algorithm, Effectively Improving Hot Data Hits
1 MB
Performance
8 KB Multi-granularity hot data statistics layer Hot data prediction based on time
64 KB Algorithm: predicts hot data sequence
Advance
prefetch
64 KB 32 KB Cold Hot
8 KB
1 MB Time
T0 T1
Current
Prefetch based on continuous

addresses Mass small files
Small I/O access
HBA address
frequency info.
Prefetch based on associated eliminated in priority
addresses
Memory
… … Small I/O
space access
HBA address
Multidimensional associated Small file accessfrequency info.
frequency info.
Elimination policies of scenario-based
Prefetch based on associated access frequency info.
times data prefetch
… … … …
Timeline
T0 T1 T2 T3 T4 Small file access

frequency info.
eliminated in priority
28 Huawei Confidential SAN/large files
50% higher in small I/Os 15% higher in small files
15000 38000
33000
9800
SAN & NAS 4500
Industry-leading performance Unit: IOPS Unit: IOPS

Test conditions: Test conditions:
Dual-controller, 48 x NL-SAS HDDs + 4 x SSDs Dual-controller, 48 x NL-SAS HDDs + 4 x SSDs
8 KB I/O size, 7:3 mixed read/write 64 KB I/O size, 6:4 mixed read/write
20% higher in large files and Equivalent with industry benchmark

small I/Os in operational I/O scenarios
42000
create
OceanStor 5510 40
35000 mkdir
Vendor H C650 30
delete
Vendor D Unity XT 680 20 access
chdir
10
Vendor N 8300
Unit: IOPS Unit: s

Test conditions: Test conditions:
Dual-controller, 48 x NL-SAS HDDs + 4 x SSDs Dual-controller, operations such as mkdir, write, read,
2 GB file size, 8 KB I/O size, 7:3 mixed read/write open, delete, access, and create
Source: internal Huawei lab data
High-Performance Warm/Hot Data Layout Powered by Intelligence: SAS

HDDs Replaced for Optical TCO
Performance
XX%
Purchase costs
Purchase costs x%
The cost per TB of SAS HDDs is five times that of
NL-SAS HDDs. Adopting the SmartAcceleration
Performance
intelligent data layout algorithm, New-Gen
OceanStor Hybrid Flash Storage can achieve
performance far higher than OceanStor V5 at a
New-Gen OceanStor Hybrid lower purchase cost, and without SAS HDDs.
OceanStor V5 Flash Storage
SSD SAS NL-SAS HDD SSD NL-SAS HDD

HDD
Interconnection Between New-Gen OceanStor Hybrid Flash Storage and

Other OceanStor Series
All-flash and cross-gen hybrid flash arrays build a complete DR system.
• The New-Gen OceanStor Hybrid Flash Storage supports SAN/NAS asynchronous replication
OceanStor Dorado V3 with OceanStor Dorado, and can work as the primary storage node of DR.
• The New-Gen OceanStor Hybrid Flash Storage supports SAN asynchronous replication with
OceanStor Dorado V3 and OceanStor V3/V5 to fully reuse legacy storage devices.
OceanStor
V3
SAN Asynchronous SAN/NAS Asynchronous

SAN/NAS HyperMetro Replication Replication
SAN/NAS Replication
OceanStor OceanStor
OceanStor Dorado All-Flash
V5 V5 kunpeng OceanStor 5310~18810 Storage
New-Gen OceanStor Hybrid Flash
Storage
SmartInsight Provides Health Analysis in Terms of System, Performance, and

Capacity
Capacity and health Performance
System health check
prediction fluctuation analysis
• Accurately identifies factors affecting • Fluctuation analysis by day, week, • Intuitively reflects the running
capacity. and month status of the equipment.
• Intelligently identifies information about • Service change guidance • Proactively prevents risks.
disk performance, logs, and features.
Performance analytics and optimization

Disk fault prediction
suggestions
Drive service life
intelligent world.


Huawei OceanProtect Backup Storage
Pre-sales Training
Contents
1. Data Protection Trends and Challenges
2. Highlights of Huawei OceanProtect Dedicated Backup Storage
3. Specifications of Huawei OceanProtect Dedicated Backup Storage
Data Is the Core of an Intelligent World and a Key Production Factor for the
Digital Economy
Core of an intelligent world Key production factor for the digital economy
32 ZB → 180 ZB Agricultural Industrial Digital

Prediction about data volume from 2020 to 2025 economy economy economy
Data Land Labor force Capital Technology Data
1 TB/day 10 PB 40 PB 64 TB/day
Production data Transaction data Analytical data Training data
Digital connected factory Huawei IT Autonomous vehicle
New Challenges to Data Backup as Industry Digitalization Grows Rapidly
Superior performance Cost reduction Security protection
Rapid data backup and recovery Ultimate data reduction Storage ransomware protection
capability capability
More than I/3 of the Data grows by more than 50% Gartner: 60% of enterprises
enterprises cannot meet each year, but the data require storage products with
backup window requirements protection budget increases ransomware protection
of mass data. less than 10%. functions.
February 2020 April 2021 July 2021 December 2021 March 2022
Core user data of an enterprise Ransomware attack on a REvil attack on a tech Data in the supercomputing Data breach of a consumer
intentionally damaged computer manufacturer management software system of a Japanese electronics company in South
Services for 3 million clients A ransomware gang called REvil supplier in the US university deleted by mistake Korea
interrupted. intruded the enterprise's More than 1 million systems About 77 TB files deleted by 150 GB of confidential data and
system and claimed a ransom locked and a ransom of USD 70 mistake, and around 3400 files core source codes stolen by a
of $50 million. million demanded. lost. ransomware gang.
The F2F2X Era Has Come, Redefining Flash-based Backup Solutions
"Disk to Disk to Tape" (D2D2T) Is Now
Out of Date F2F2X
Flash-to-Flash-to-Anything
Primary Backup Archive &

Storage Storage Other
Copies
Longer backup window, lack of efficiency towards
Anything: Public/Private/Hybrid
new scenarios Already flash-based To be flash-based
cloud, disks and tapes, and blu-ray
Long and non-predictable RTO
Weak data reduction capability, failure of tape Backup Tier/Archive

data deduplication, and consumption of massive
storage space
Restart the backup job in the mid-night Performance Cost System resilience
Bandwidth improved by TCO reduced by 20% Disk return rate decreased from
3 to 5 times 5% to 0.3%
Contents
OceanProtect Backup Storage Leads Backup into the F2F2X Era
Oracle VMware SQL

OceanProtect
Linux SAP … Backup Storage
(large capacity)
Long-term
Backup media Backup media archiving
server server Backup server
Object storage
Production host Data backup
• Sharp increase in data Tape library

volume Remote
replication
• Instant recovery
• Diversified backup
objects
OceanProtect OceanProtect Cloud
Backup Storage Backup Storage
All-flash storage
Backup environment
Primary Backup Archiving

Storage Storage
Disks, tapes, and clouds

All-flash-ready Going all-flash
Diversified data copies
OceanProtect Backup Storage: The Enterprise-Class Industry Benchmark
Rapid Backup, Rapid Recovery

up to 172 TB/hour recovery for ultimate
performance, ensuring high data availability.
…
Backup media Backup
servers server
Efficient Reduction
IP/FC
72:1 data reduction ratio using leading
algorithms, greatly improving effective
capacity and reducing system investment.
Solid Resilience
Industry-leading ransomware protection solution
OceanProtect Backup Storage
and exclusive active-active architecture, ensuring
the secure and reliable data copies.
Certification of Huawei OceanProtect by Authoritative Organizations:
Evaluator Group and Interop Tokyo
Evaluator Group Interop Tokyo Award
Evaluator Group, a leading IT analyst firm Authoritative awards in the Storage & Server Categories
worldwide, released the technical review report on
Huawei OceanProtect Backup Storage. Following
rigorous tests on the live network, Evaluator
concluded that
"Huawei OceanProtect delivered more than 2X faster

backup data rates than its leading competitor. In many
cases, protecting data 2X faster means being able to use
1/2 as many systems as the competitor... Huawei
OceanProtect also has the ability to expand to two nodes
for load balancing business, creating additional opportunity
for improved system utilization."
"Huawei OceanProtect provides significant capacity

benefits vs. the competitor. By achieving approximately
50% greater reduction ratios for various data, Huawei
OceanProtect has an effective capacity of nearly 50%
greater than the leading competitor when configured with
the same raw capacity."
The largest and most influential
The well-known IT analysis organization: ICT exhibition in Japan: Selects
Demonstrates vendors' leading technologies
out the most competitive
through real-world review and provides
professional and objective recommendations products in the industry.
Rapid Backup Efficient Solid
Rapid Recovery Reduction Resilience
Flash-based E2E Acceleration Improves System Performance

Front-end network optimization Shortens backup windows
DTOE: Full protocol
stack offload
I/O
24
PHY 1 PB
MAC hours
IP
DTOE NIC
Backup storage products in the
TCP industry (40+ TB/hour)
➢ Industry-leading 10/25/40/100GE OS
Buffer Driver
Backup data 6 hours
Protocol User
➢ TCP/IP protocol parsing to release CPU FileSystem
space OceanProtect (155 TB/hour)

resources
Backup speed up to 155 TB/hour and

OceanProtect Backup Storage 10 Gbit/s, 25 Gbit/s,
restoration speed up to 172 TB/hour
40 Gbit/s, or 100 CPU computing
Gbit/s DTOE optimization
iNIC iNIC
Name Description Quantity
➢ Intelligent CPU partitioning and grouping
algorithm OceanProtect X9000 OceanProtect X9000 2
Controller Controller ➢ CPU multi-core load balancing optimization controller enclosure storage controller
enclosure A enclosure B ➢ Sequential large I/O optimization for backup enclosure (four
services
controllers)
Number of disk 2 U 25-slot 2.5-inch 4
Back-end network
enclosures per node SAS disk enclosure
Disk Storage Pool Disk optimization
enclosure 1 enclosure 1 ➢ High-performance SSDs and hardware Number of SSDs per 3.84 TB SAS SSDs 75
acceleration to reduce I/O latency node
Disk Flash- Disk ➢ ROW full-stripe write to reduce write
enclosure 2 based enclosure 2 amplification
➢ Adaptive data locality rearrangement on writes
* Test results in Huawei lab using the Smartbench tool.
Three-Layer Metadata Structure and Adaptive Data Locality Rearrangement

Improve HDD Recovery Performance
Traditional structure Locality rearrangement (among 4 MB data blocks)
LUN map Read and write in the LBA order. HDD-based recovery
Data locality rearrangement effect
performance: 20% better than
Backup data is read and written in logical order. peer product
Fingerprints are randomly
reorganized by content.
Fingerprint A B C D E F G H I Overall leading recovery
table
bandwidth
The data sequence on the disk is consistent with
the fingerprint sequence.
Single-node recovery bandwidth
+
On-disk data
Data locality Unit: TB/hour
rearrangement
24
Standard backup and
Three-layer metadata structure dedupe process 18
Read and write in the 12

LUN Map LBA order Duplicate check table
8 8
Actual storage
A B
C EC D
H E
G BF GI H
D 1
F location
6
Points to old IDs for Assigns new IDs for new
duplicate data. data
...
ID index table
...
On-disk read order
The ID sequence is the consistent with the 1 3 5 8 7 2 9 4 69
1 2 3 4 5 6 7 8 Mid-range Lower high-end High-end flagship
LBA sequence.
Points to historical data blocks
Writes new data to disks On-disk data
for duplicate data. The actual storage location of data on disks is the Peer HDD-based storage
... ...... layout Huawei HDD-based storage
same as the logical sequence.
The data sequence on the disk is consistent with the
ID sequence.
Data is randomly stored on disks.
4-Step Advanced Deduplication and Compression for Optimal Data Reduction

Periodic Chunking rule 2
backup Backup data Data Non-duplicate data
Chunking rule 1
deduplication Different
application data
A B C
X
D E F
Structured Changed Multi-layer variable- Fingerprint calculation, Write data to disks
data head data length segmentation search, and comparison after compression
Dedicated compression
algorithm for backup
① Backup data ② Multi-layer inline variable- ③ Feature-based ④ Byte-level

preprocessing length dedupe compression compaction
Data volume after

Up to 72:1* Data volume before reduction reduction
data reduction ratio
* Test case: 200 VMs in a virtualization scenario, daily full backup for 28 days, 4% data changes per day, 0.125% new data per day
Data Reduction: Backup Data Preprocessing Separates Data Heads from

Changed Data
Production Historical Changed
data
environment file
Backup data
Data head Changed
generation
Backup server data
The data head contains random data such as

compression and parity data.
Backup data
Traditional backup preprocessing OceanProtect
storage Backup Storage
Whole backup data

dedupe
Data heads are separated
from changed data.
Data to be deduplicated contains the data heads, reducing

Variable-length chunking dedupe is
the overall deduplication ratio.
performed after aggregation of changed
data blocks.
OceanProtect can preprocess multiple types of backup data, such as Oracle RMAN and Commvault
* Vendor E supports only the preprocessing of Oracle RMAN backup data. * backup data.
Data Reduction: Multi-Layer Inline Variable-Length Deduplication of Backup

Data Improves the Deduplication Ratio
Multi-layer inline variable-length
Traditional variable-length deduplication
deduplication (Huawei-patented)
Variable-length deduplication uses a sliding window to scan I am a boy, my father is james.
backup data to find a proper chunking point that meets Chunking for I am a boy, my father is james.
preset conditions. multiple times
I am a girl, my father is james.
Two preset conditions:
• Use a period (.) as the segmentation point.
Backup data I am a boy, my father is james.
• Use a comma (,) as the segmentation point.
I am a boy, my father is james. Data blocks to be
Segmenting once deduplicated
Preset conditions: I am a girl, my father is james.
Use a period (.) as the segmentation point. Dedupe ratio
10%
I am a boy, my father is james.
Data blocks I am a boy, my father is james.
I am a boy, my father is james. after deduplication I am a girl, my father is james.
Data blocks after
I am a boy, my father is james.
deduplication
I am a girl, my father is james.
Chunking only Deduplication Chunking for Deduplication

once ratio 33.33% multiple times ratio 50%
There is a low probability that proper chunking points are found. As a Find proper chunking points, thereby improving
result, the deduplication ratio is low. the data deduplication ratio.
Data Reduction: Feature-based Compression of Backup Data Greatly

Improves the Data Compression Ratio
Huawei's feature-based compression
algorithm improves the compression ratio
by 57%.
Compression ratio
Data block 1B 8B
combination 4.5
(larger compression field) A B C D a b c d 4
3.5
All letters are changed to
3
uppercase letters.
Data feature 3 bits 1B 8B 2.5
preprocessing 2
(improved data compressibility) A B C D A B C D 1.5
Metadata 1
tag
Byte-level dedupe and 0.5
compression 3 bits 1B 4B 0
(byte-level de-redundancy)
A B C D LZ4 GZ fast GZ HZBC
Metadata
tag
High-performance predictive Feature-based compression: The Huawei's patented
entropy encoding
(bit-level redundancy removal)
HZBC algorithm provides a leading compression ratio
which is 57% higher than LZ4, 27% higher than GZ fast,
and 14% higher than GZ.
Storage Deduplicated Replication Helps Greatly Save Replication Network

Bandwidth
Lower TCO
Data Center 1 Data Center 2
• With deduplication for replication, non-duplicate data is
transmitted, saving bandwidth.
Backup server Backup media
Backup server Backup media
server
• With replication data compression, the amount of data to be
server
transmitted and TCO are reduced.
• Supports interoperability and replication between different
product models.
• Data can be backed up and replicated at the same time.
①
Shorter replication duration

OceanProtect backup storage OceanProtect backup storage
③ Replication performance 10 times that of peer vendors
④
File system Storage File system Backup OceanProtect storage replication
Backup Replication bandwidth
images replication images
6000+
②
Oracle Windows file Unit: MB/s
Application Swift
(RMAN file) system
Replication data volume

(not for the initial 59.7 GB 83.3 GB 78.9 GB
1 Reads fingerprints of data and sends 2 The target device checks duplicate data replication)
590+
them to the target device. and returns the result. Replication bandwidth 6.63 GB/s 6.94 GB/s 6.57 GB/s
3 Packages and sends non-duplicate data 4 Writes data Replication time 9s 12s 12s Vendor N OceanProtect
(compression available). * Project test environment
OceanProtect X8000, NetBackup, and two data centers are 500+ km away from each other.
* Network requirements:
• RTT ≤ 100ms
• Bandwidth ≥ Changed data volume during peak hours/Replication period
• Packet loss rate < 0.1%
Source-side Deduplication Technology Based on the DataTurbo Client

Greatly Reduces Backup Time
Lower TCO
Backup Backup media Backup media Backup media • DataTurbo client should be deployed on the backup media
server server server server server.
• Source-side deduplication reduces the amount of data to
①③ be transmitted and saves network bandwidth.
DataTurbo DataTurbo DataTurbo • Source-side deduplication and transmission of non-
duplicate data improve logical backup bandwidth.
② IP/FC IP/FC IP/FC
Shorter backup windows

OceanProtect backup storage
10 hours
1 PB
File system Backup
images
④ Backup storage boost in the
industry (90+ TB/hour)
Segments data and Compresses non-duplicate data and

1 3 packages and sends them
Backup
calculates fingerprints data 3 hours
2 Checks duplicate fingerprints and 4 Writes data DataTurbo (310 TB/hour)
returns the results
* OS compatibility: CentOS 7.x, SUSE12 SP4/SP5, Ubuntu 20.4, Oracle Linux 7.6, and more OSs are supported. For more information, visit the compatibility website at
https://info.support.huawei.com/storage/comp/#/oceanprotect.
* Network requirements: RTT ≤ 100ms, packet loss rate < 0.1%
* Network type: IPv6 is not supported.
Industry-leading Ransomware Protection Solution Ensures Security and

Availability of Copies
Production data center Backup system Backup isolation zone
Base protection Advanced protection
Local protection of the Local protection of the backup system + Air
backup system Gap + Replication link encryption + Isolated
environment protection
Data leakage Data tampering Data leakage Data tampering

prevention prevention prevention prevention
Storage encryption, Secure snapshot Storage Secure
protocol encryption WORM file system Physical encryption snapshot
isolation
Air Gap
Restores local Data
Replication link
secure data. restoration
encryption
Production data Protocol encryption
Restores secure data from
Backup data flow the isolation zone.
Restores secure data from the isolation
zone to the production zone.
1 2 3 4 5 6
Replication link
technologies
Protocol encryption Array encryption Secure snapshot WORM Air Gap

encryption
NFS, CIFS or SMB packets are It ensures the security of The backup storage in the It prevents data written in the Automatically disconnects the
During replication,
Key
transmitted in ciphertext to stored data and addresses the production center and security file system from being replication link to copy data
transmission encryption
prevent data from being risk of sensitive information isolation zone uses the secure modified or deleted, or to an isolation area for higher
ensures data security and
stolen. leakage caused by snapshot technology to ensure encrypted by ransomware. security.
prevents sensitive
ransomware attacks. that the snapshot data is read-
information leakage.
only and cannot be modified or
deleted within a specified
period.
Active-Active Architecture and RAID-TP Tech Ensure System-Level Reliability
Active-Active RAID-TP
Controller Controller
Tolerance of simultaneous 3-disk failures

without service interruption
Non-disruptive service switchover

If a controller fault occurs, services can be failed over within
seconds without interrupting backup services.
OceanProtect X9000 tolerates failures of three out of four
controllers.
DME IQ Shifts O&M from Reactive to Proactive to Ensure Stable Service

Running
Mobile O&M Intelligent prediction Intelligent analysis
• Access using mobile apps • Capacity prediction (one year in

advance) • Performance bottleneck analysis
• Automatic creation of maintenance
service tickets • Disk fault prediction (14 days in
advance)
Mobile app: Allows users

to view and manage
Huawei storage systems at
any time
Disk fault prediction Performance analysis and
optimization suggestions
Contents
Product Overview High-end OceanProtect X9000
Entry-level OceanProtect X3000 Mid-range OceanProtect X6000 High-end OceanProtect X8000
New
OceanProtect X3000 OceanProtect X6000 OceanProtect X6000 OceanProtect X8000 OceanProtect X8000 OceanProtect X9000 OceanProtect X9000
(HDD mode) (all-flash mode) (HDD mode) (all-flash mode) (HDD mode) (all-flash mode) (HDD mode)
Single-node
2 U, 2 controllers 2 U, 2 controllers 2 U, 2 controllers 2 U, 2 controllers 2 U, 2 controllers 4 U, 4 controllers 4 U, 4 controllers
specifications
Maximum number of
1 1 1 2 2 2 2
nodes*
Data disk type NL-SAS HDD SAS SSD NL-SAS HDD SAS SSD NL-SAS HDD SAS SSD NL-SAS HDD
Capacity per data 4 TB/8 TB/14 TB NL- 4 TB/8 TB/14 TB NL- 4 TB/8 TB/14 TB NL-
3.84 TB/7.68 TB 3.84 TB/7.68 TB 7.68 TB 8 TB/14 TB NL-SAS
disk** SAS SAS SAS
System usable
16 TB to 60 TB 16 TB to 300 TB 150 TB to 2.0 PB 480 TB to 3.6 PB
capacity
System backup
Up to 6 TB/hour Up to 19 TB/hour Up to 19 TB/hour Up to 55 TB/hour Up to 55 TB/hour Up to 155 TB/hour Up to 155 TB/hour
bandwidth
System logical backup
bandwidth
System restore
bandwidth
Front-end port type 8/16/32 Gbit/s FC, 10/25/40/100GE
Back-end port type SAS 3.0
Note: *: A node corresponds to a controller enclosure.

**: 22 Huawei
If disks of otherConfidential
capacity specifications are required, contact Huawei sales personnel for evaluation.
Standard NAS/SAN Protocols Are Compatible with Mainstream Backup
Software
Production Mainstream backup system
system
Commvault Veritas IBM TSM Networker …
Oracle
SQL Server
CIFS/NFS/FC SAN/IP SAN
MySQL
MongoDB
Hadoop
OceanProtect X8000 OceanProtect X9000 OceanProtect X6000

(controller enclosure (independent disk and (controller enclosure
integrated with disk slots) controller enclosures) integrated with disk slots)
intelligent world.


Huawei OceanStor Pacific Series Pre-sales
Training
Security level:
Contents
1. Distributed Storage Concepts
2. Highlights of Huawei OceanStor Pacific Distributed Storage
3. Sales Scenarios of Huawei OceanStor Pacific Distributed Storage
Distributed Storage Becomes Increasingly Essential
Storage Growth Extensive

applications
HCI
Cloud transformation
Distributed Banking, ISPs
Device and data integration

All-flash Carriers, large enterprises, electric power
industry
Video networking
Converged
Government, law enforcement, public safety,
campus services
2016 2017 2018 2019 2020 2021 2022

Source: HUAWEI MI
Distributed Storage vs. Centralized Storage
Distributed Storage Centralized Storage
Cluster + Identical Nodes Controller + Disk Enclosure
Sever Sefver Server Server
VM VM VM VM DB VM VM
VM VM VM VM 1. Larger capacity: VM VM
1. Smaller capacity：
XXX TB~EB TB~PB
2. Elastic Expansion Switch

2. Limited expansion:
switch
with adding nodes, 2~XX controllers
up to thousands.
SAN/NAS Array 3. Performance: IOPS，
Node 1 Node 2 Node N 3. Performance：
Controller XX K ~ X M IOPS
… bandwidth XX
GB/S DAE DAE DAE
NLSAS SAS SSD
Distributed Storage
XX controllers
Hundreds to thousands of nodes
Linear expansion of capacity and performance
Application for Distributed Storage: ①diversified data ②Massive expansion ③ simplified management
Catalog of Distributed Storage
File HDFS
Block
Video editting • Works as traditional SAN
Logs
HPDA • Application: low latency, high IOPS
Analysis
CCTV
bigdata
PACS
Block Object
File
• Works as shared folder
VM Image • Application: File sharing, HPC,
Video, Image
Database Online video
Container Backup & Archive Object
• Internet access
• Application: Internet App,
SCSI/iSCSI NFS/CIFS S3 HDFS IOT
HDFS
• Similar as file
• Application: bigdata platform
Distributed Storage
Contents
Huawei OceanStor Embraces the Era of Mass Data with Continuous
New
Innovations OceanStor Pacific scale-out storage
Gartner Peer Insights Customers' Choice 2022
OceanStor Pacific next-gen HPDA parallel storage

Best of Show Award Grand Prize at Interop 2021
OceanStor Pacific hybrid version released

The only Chinese vendor that enters the Gartner Challengers quadrant 2020
Best of Show Award Grand Prize at Interop 2019
Industry's first distributed active-active solution 2018
No.1 market share in China, ranked by IDC 2017
Distributed block storage

No. 1 performance in SPC-1 2016
Pioneer in commercial use for 4K UHD 2015

2nd-gen file storage
OPS: a new record in SPEC test 2013
1st-gen file storage 2009
Huawei Distributed Storage Products Have Gained Industry Recognition via
Continuous Upgrades
New
Best of Show Award Grand Prize at Interop One of the world's leading HPC storage performance
Tokyo 2021 rankings Gartner Peer Insights Customers'
Choice 2022
3rd place worldwide (1st place as a commercial
product)
OceanStor Pacific Series Overview
High-performance Content resource pool Log retention Smart city Virtualization/Cloud
resource pool Database
computing Backup &
Application Web disk
archiving
layer
NFS/SMB
S3 HDFS SCSI/iSCSI
/POSIX/MPI-IO
SmartMulti- Self-encrypting drive

SmartTier SmartQuota SmartQoS SmartEqualizer SmartThin SmartQoS
Tenant (SED)
layer
HyperReplication
(S)/(A)
EC/Multi-copy Fast reconstruction SSD cache DIF

..
.
OceanStor OceanStor Pacific 9350 OceanStor OceanStor Pacific 9550 Pacific 9950
Products Pacific 9340 Pacific 9346 OceanStor Pacific 9540 Pacific 9546 OceanStor
Pacific 9920
Pacific 9520
OceanStor Pacific Product Overview
Decoupled storage- Backup and
HPDA compute for big Virtualization/cloud
Intelligent video & image archiving of
data resource pool
mass data
Balanced Archive
OceanStor OceanStor Pacific 9550
Performance OceanStor Pacific 9546 OceanStor Pacific 9340 OceanStor Pacific 9350
Pacific 9950
OceanStor Pacific
9920
OceanStor Pacific scale-out storage

Specifications
Performance model: 99 series Balanced model: 95 series Video model: 93 series
OceanStor Pacific 9950 OceanStor Pacific 9920 OceanStor Pacific 9350
Product Model Pacific 9546 Pacific 9550 Pacific 9346
Configuration
(80 slots per enclosure) (25 slots per enclosure) (120 slots per enclosure)
Hardware
(60 slots per enclosure) (120 slots per enclosure) (60 slots per enclosure)
Main Storage Disk Type NVMe SSD SATA HDD SATA HDD
Dedicated hardware Dedicated hardware Dedicated hardware Dedicated hardware Dedicated hardware Dedicated hardware
Chassis Model
5 U 8-node 2 U 1-node 4 U 2-node 5 U 2-node 4 U 1-node 5 U 2-node
(Key Features of Version 8.1.5)
Storage File, object, block, and HDFS File, object, block, and HDFS File and object
NFS over RDMA Supported Supported Supported Not supported
Ransomware Protection Supported Supported Supported Supported
Functions
Antivirus Protection Supported Supported Supported Supported

Intra-Cluster Data Tiering and
Supported Supported Supported Not supported
Mobility
Inter-Cluster Data Tiering and
Supported Supported Supported Supported
Mobility
OceanStor Pacific : Massive Data Storage Platform for Hybrid Loads
HPDA BigData Video & Imagine Backup & Archive
OceanStor Pacific
Performance Balanced Video

OceanStor Pacific OceanStor Pacific OceanStor Pacific
9950 | 9920 9550 | 9546 9350 | 9346
High efficiency with Hybrid loads Optimal TCO for Massive Data High Data Resilience
• SmartInterworking: Industry-Leading, • High-density hardware: 120 HDD/5U, saving • HyperGeoMetro & EC：Multiple-DC DR
Lossless Interworking Between Unstructured space and power. solution, supporting up to 12 DCs.
Protocols
• Automatic inter-intra Data Flow: built-in data • Fast Failover: failover <10S
• One nodes supports multiple-protocol services. tier make data migration more smoothly and
cost-effective. • Data Security：Built-in intelligent
ransomware detection, with a detection and
• Excellent Expansion Capability：Fully identification rate of 99.9%, industry-leading,
symmetric distributed architecture, linear comprehensive protection to ensure high
growth of performance and capacity. data security
Always cost-
Always efficient Always on
effective
SmartInterworking- zero data copy, instant read and write of different

applications
Traditional solution: Separate file and object indexes Huawei Solution: Unified File and Object Index
Data
Data import Training & Simulation Data release Data import Data preprocessing
Training & SimulationData release
preprocessing
Unique in
the industry
S3
S3
NFS/CIFS
Copying Copying Copying space
50%
File Big data File Object File Big data Object
0 gateway, 0 loss, 0 copy
Analysis efficiency
30%
Raw data Preprocessing OPS storage OBS OceanStor Pacific

library
Bandwidth-based storage
Always cost-
Always efficient Always on
effective
Hybrid load performance balancing: I/O data flow adaptation
Large I/O Small Large I/Os are directly written to the storage tier.
I/O 1
Reduces network and memory bandwidth amplification and
improves bandwidth.
1 Small I/O cache tiers write after aggregation

Full 2 Reduces the number of I/O interactions, reduces write penalty, and improves
Cache random small I/O write performance.
Stripe
layer
2 3 3
Intelligently prefetch files
1 2 3 4 5 6
Storage 3 Sequential small I/O read for large files and intelligent prefetch for small file
Large File Small
layer indexes
Sequential small I/O File
160 GB/s 6.4 million IOPS 640K OPS
Single frame
Always efficient Always cost-effective Always on
Optimal High-Density Hardware: Easy Installation and Maximum Value of Data

Center Space
High-density 20% 14.5%

5U 120 design Hard disk density per
unit space
Large EC
Space utilization
Subrack 3.5-inch SATA
hard drive
91.6%
OceanStor Pacific 9550
80%
5 U, 120 disks
4 U, 80 disks
Other vendors Huawei Other vendors Huawei
• Up to 22+2 EC
• Flexible number of nodes
Note: The space usage does not exclude the reserved space.
Tiering and Free Flow of Hot, Warm, and Cold Data, Reducing TCO by 70%
Real-time service Hybrid services Low-frequency services
≥
70% TCO reduction *
AWS
Huawei Cloud • Automatic data tiering: supports free flow
...
combination and
combination and compression

Tiering of hot, warm, and cold data from one
compression
Tiering
Multi-model
SSD HDD Data Cloud cluster to multiple clusters or clouds at
Tiering
multiple data centers, with customizable
Multi-model
OceanStor Pacific OceanStor Pacific HDD
Data
9950 | 9920 9550 | 9546 data tiering policies.
OceanStor Pacific
Blu-ray disc library • Data compression before flow: offers a
Blu-ray
maximum compression ratio of 2:1,
Tiering exceeding the industry by 40%, reducing
OceanStor
OceanStor Dorado OceanProtect
self-developed storage space and bandwidth.
Blu-ray storage
* Accounting under typical configurations may

Enterprise multiple data centers vary in different scenarios.
Optimal linear expansion: 100 billion objects in a single bucket, meeting the
requirements for efficient access of massive small files
The number of objects in a
single bucket is far ahead.
e-bill image Checkpoint image IoT Internet of
Vehicles
15M 1000M 1000M 100,000M
Vendor X Vendor E Vendor H HUAWEI
Stable performance
• 100 billion-level • A large city, 2W traffic • Large object,
TPS
objects in a bucket cameras, 140 billion bandwidth > 250 MB/s
Third-party test data and 10 billion-level images in a year • Small object, TPS of a
HUAWEI
1,000,000 objects in a single set single bucket: 300–
• Performance: TPS 600
1000 to 10000
Vendor X
1,000,000
100,000 million
Meets the single-bucket read and write requirements of service applications, eliminating bucket reconstruction for large-scale
applications.
Always-on Services: Cross-site DR + Second-level Switchover upon nodes

fault within a site
Set of Data
A B C D P P
Site-based active- ≤10s

12 active DR (object)
Node Failover
• Supports 2 to 12 sites in multi- • Intelligent heartbeat detection

active mode.
• Quick service takeover
• A maximum of two sites can be Site 2
• A maximum of four nodes fail at
faulty at the same time.
the same time and services are
• Industry-leading space Intra-site storage cluster
not interrupted.
utilization: For example, the
Site 1 Multi-site multi-active Site 12
utilization rate reaches 61% in
the case of three sites in multi-
active mode.
Site 3
Cross-Site DR Demonstration
HyperGeoMetro HyperGeoEC
Site 1
Site 2
Site 1 Site 2
Bucket A
Bucket A
A B
C P
HyperGeoMetro WAN Bucket D
Bucket A Bucket A
Site 4 Site 7
File
File11 HyperGeoMetro File 1
Bucket B Bucket B Bucket A
Bucket A
Host 1 …
Bucket D Bucket C
Site 3
File 1
File 2
Site 12 Site 5
Host 1 Site 12
…
File
File 22
Bucket B
Site 5
Bucket C
Always-on Services: Built-in Detection of Ransom, Comprehensive Protection

Production Site Isolated Site
Foundation protection Advanced protection

Local protection for production storage + Air gap + replication links
Local protection for production storage Encryption of + Air-Gap protection
Data leakage Data Tamper Detection and Data leakage Data Tamper Detection and
prevention Protection analysis prevention Protection analysis
Storage Encryption Secure Snapshot Pre-mortem detection
Protocol Encryption WORM Perimortem detection Storage Encryption Secure Snapshot post-event detection
Physical
post-event detection
isolation
Air Gap
Replication Link Data recovery
Local Secure Encryption
Data (2023H2)
Recovery Quarantine
Restore security data from the Security Data
OceanStor Pacific Air-Gap area to production OceanStor Pacific Recovery
Key technologies
1 Detection and analysis 2 Secure Snapshot 3 AIR GAP 4 Replication Link Encryption 5 Protocol Encryption 6 Storage Encryption
Huawei-developed detection and Storage in the production center Data is copied to the isolated During the replication process, Ensure the security of stored data
NFS and CIFS/SMB
analysis engine is installed. It and Air-Gap area uses the secure area by automatically shutting transmission encryption and prevent the risk of sensitive
packets are transmitted
analyzes file I/O behavior and file snapshot technology to ensure down the replication link, which ensures data transmission information leakage caused by
in ciphertext, preventing ransomware attacks.
damage characteristics based on ML that data is read-only and cannot provides a more secure security and prevents sensitive
data theft.
algorithm, generates event alarms, be modified or deleted within the protection effect. information from being (2023H2)
specified time range. (2023H2) disclosed.
and provides client information for
source tracing.
Contents
OceanStor Pacific Scale-Out Storage: One Storage System for Multiple
Workloads
Four Scenarios
Scientific research HPDA Video and media Mass data archiving Medical imaging center
• Education: HPC in higher education • Broadcasting and TV: HD editing • Finance: financial document imaging • Healthcare: PACS system
institutions
• Government: safe city • Education: data archiving • Healthcare: digital pathology
• Government: supercomputing, scientific
research, and meteorological HPC • Transportation: highway, airport, • Government: data archiving
• Healthcare: genome sequencing and urban rail
• Manufacturing: autonomous driving and oil
and gas exploration
Huawei OceanStor Pacific - next-generation distributed storage system that supports large-scale scale-out and
hybrid loads
Performance Models Balanced Models Video Models Archiving Models

OceanStor Pacific 9950 | 9920 OceanStor Pacific 9550 | 9546 OceanStor Pacific 9350 | 9346 OceanStor Pacific 9150 | Huawei-developed
(2023H2) Blu-ray storage
(2024 H1)
HPDA-One Storage Device Meets Multiple Service Requirements and
Delivers Leading Performance in All Scenarios
Energy
exploration
Supercomputi
ng Center
Gene
sequencing
educational and
scientific research
Autopilot 7mins
Clinical Medicine ESI Top One Thousand in the World
HPC cluster/GPU workstation AI/Big Data cluster Archive management server
data center network InfiniBand/RoCE hereditary population Assisted Oncology contagious ICU early sieve
POSIX, MPI-IO, NFS, S3 disease cohort study reproducti Research disease
research on
Production storage Archive storage

OceanStor Pacific
Human genome-wide analysis enters the minute-level era for
Intelligent data tiering and the first time
compression
OceanStor Pacific 9950 OceanStor Pacific 9550 | 9546
High-density performance High-density balanced 30X Human WGS Germline Variation Analysis Time
One set of storage devices open 24 hours

source
Concurrent interconnection software
Industry No.1 in Multiple Lossless multiple-protocol
of diverse computing Performance Models interworking FPGA 25
power acceleration minutes
WHS-IMOAP
Native objects, files, and big data 7
ARM+AI+X86+GPU 160 GB/s bandwidth, 6.4 million IOPS 2.0 minutes
Single-stream bandwidth: 8 GB/s Multi-Protocol Interworking
Data Backup and Archiving Solution-Efficient, Reliable, and Cost-effective
Object Data Platform
Internet
Efficient Top Banks
Content Stable and high performance in the
Ticket image Video of
collaboration case of 100 billion object storage
Things
Ticket image
TPS 7 x 24 service continuity 4x faster processing speed
Third-party test data
assurance
Atempo software 1,000,000 Huawei
PSM
Rubrik
... Product C
Content management 1,000,000 100,000 million
platform
electronic statement
system
Reliable
S3 Cross-site DR supports a maximum e-bill ......
Global
Namespace of 12 sites in multi-active mode.
Industry-leading reliability and
... space utilization
Legacy system: Active-
OceanStor Pacific Legacy file Active
Site 1 Site 12 OceanStor Solution OceanStor
system
economy Pacific Pacific
Large EC ratio provides optimal site1 Site 2
Site 2 resource utilization.
Intelligent Video and Image Storage-New-Generation Stream Forwarding &
Storage for Video Security, Responding to the Video Explosion Era with Ease
Video security service flow
Multi-protocol
More flexible scalability
Fully symmetric distributed architecture: up to
interworking
4096 nodes for a single cluster, with no
Seamless NFS/CIFS/HDFS/S3 interworking with
independent
zero metadata
data copy helps node easily
our customers
Safe City Urban rail Airport Subway implement HPDA- and AI-oriented
Efficient management evolution.
and easy deployment:
efficient, easy-to-use, and unified cluster
management mode
Video/Image
VMS
Video/Image
Access Multi-protocol
Lower cost
interworking
High-density hardware: 4 U/5 U high-density
Seamless NFS/CIFS/HDFS/S3
models save equipment interworking
room space, with
cabinet
zero data copy helps our customers easily
TCP/IP power consumption, and total cost of
AI analysis implement HPDA- and AI-oriented evolution.
operation (TCO).
Large-ratio EC: 22+2
NFS/CIFS
Multi-protocol
Higher reliability
Camera video
stream OceanStor Pacific series stream interworking
System-level high reliability: 99.999%
Camera image Storage reliability, dynamic EC, and
Seamless NFS/CIFS/HDFS/S3 tolerance
interworking of
with
stream
forwarding and storage base zero data copy helps
four faulty nodes our customers easily
implement
Service HPDA- and assurance:
continuity AI-oriented evolution.
failover in
seconds, and online upgrade with no
service interruption
Media Service Storage-Cost Effective, Performant, and Streamlined Storage
for HD Video
Storage Solution
Collection Editing Management Archive Broadcast
Satellite/Filming/DVD Editing/Rendering/Sy
nthesis
Cataloging/Indexing/T
ranscoding
CD-ROM/Storage Broadcast and control Unique 36-layer
4K UHD editing
…
Collection Resource Long-term

NLE content
system data scheduling archive (S3/NAS) All-day playout
production
input (FTP/NAS)
4,096 nodes
horizontal scale-out
Unified converged resource pool 24/7 playout
scalability
Performance Models Performance layer Capacity layer Archive layer Cloud All-flash storage
OceanStor Pacific 9950/9920 Tiering AWS (Fault domain isolation)
Auto migration Auto migration
Huawei Cloud
Balanced Models OceanStor Pacific 99xx
Etc.
100%
OceanStor Pacific 9550/9546
OceanStor Pacific 99xx OceanStor Pacific 95xx OceanStor Pacific 91xx data security
Archiving Models
Features of the storage resource pool
OceanStor Blu-ray storage (2024H1)
High-density hardware design | Fully symmetric horizontal scale-out architecture | Intelligent prefetch
algorithm | FlashLink | Intelligent data tiering | Ransomware protection | Recycle bin | And more
Bring digital to every person, home
home, and
and
intelligent world.


Huawei eDesigner & SCT Tools
Pre-sales Training
Can't finish product evaluation or
product configuration in several hours?
⚫ Can master complicated calculating methods
⚫ Used to looking up massive amounts of data

Can't guarantee
⚫ Have manual computing expertise
the accuracy?
⚫ Familiar with product specifications
⚫ Rely on own experience
Even if we have the skills mentioned

above, we may still have problems.
Takes only 10 minutes
Just input performance requirement
or device configuration into
eDesigner
• Takes just a few minutes to generate a final

solution
• Designed for Quick learning
• Professionally designed by experts in

relevant fields
Contents
1. About the UniSTAR SCT and eDesigner

2. Designing Solutions
3. Placing Orders, and Sharing Solutions
4. Obtaining Help
Placing an Order: What Is SCT?
• The UniSTAR SCT is a new generation
configuration and quotation tool that adopts
the B/S architecture.
• In the bidding and contract signing stages,
product managers can use the UniSTAR SCT for
performing configuration and providing
quotations. Project data can be used in
manufacturing, shipment, project preparation,
and project implementation stages until the
contract is complete.
Placing an Order: SCT Materials
This picture is SCT homepage, you can find the user guide and video at the right corner and click it.
What Is an eDesigner?
eDesigner introduction
eDesigner, an online Huawei solution design tool, is
developed for product managers, channel partners, and SAs
to conduct project expansion and bidding.
Online solutions
1. Storage eDesigner classification:
✓ All-Flash Storage
✓ Scale-Out Storage
✓ Data Protection
✓ Virtualization
2. Features:
✓ Integrated
✓ Professional
✓ Flexible
✓ Simple
Application Scenarios
User Roles
Product managers, channel partners, and SAs
Application Scenarios
Market exploitation Project bidding

✓ Professional platform and fast solution generation ✓ eDesigner helps create a fast solution design, greatly
improve customers' loyalty. improving solution configuration efficiency.
✓ Customers can design solutions and display ✓ eDesigner automatically generates solution technical
solutions and products, improving customers' proposals and configuration lists, serving as comprehensive
experiences. and useful bidding materials.
✓ eDesigner connects to the UniSTAR SCT with a single click,

achieving fast configuration and generating quick order
issuance.
Accessing eDesigner (1): Navigation Path
The eDesigner of enterprise has been integrated into the SCT platform, you can visit by link, and then click “Solution”.
You can also access it using the following methods.
Search for Huawei SCT in Google, and then click “Solution” and “Data Storage”
Accessing eDesigner (2): Select Location
After accessing eDesigner, select the country or area where you are located. After you select a country, eDesigner
automatically memorizes your selection.
Click here to change your location
Select a country or region
Accessing eDesigner (3): Registered User
Registered user: Use all the functions of the eDesigner platform
Click User Login to log in as a

registered user
Note:
The eDesigner platform is a free
solution design tool and can be visited
online by Uniportal or W3 users.
Recommend viewing in 1,440 x 900

resolution
Recommended browsers: Chrome
56.0 and IE11, or the latest versions
Applying for Permission
Huawei-certified partner
Employees of Huawei's channel partner Administrators of Huawei's channel partner
1. Log in to Huawei’s official website and apply for Uniportal 1. Approve the applications for associating organizations on the
user account. UIPM.
2. Use your Uniportal account to log in and apply, so you can 2. Submit the applications of applying for eDesigner permissions
start associating with the Huawei channel partners organization for the employees on the UIPM platform.
at the UIPM platform.
Contents
4. Obtaining Help
Solution Design Process
eDesigner provides wizard-based design steps, product Capability Evaluation and Integrated
Solution design. You will complete the solution design in 3 steps for Configuration Wizard.
The following uses the Storage Configuration Wizard (OceanStor Dorado) solution as an
example.
01 02 03
Design Product Design Finish Design
Solution Service Solution
Support Solutions
eDesigner supports product solution of Datacom, Optical, Data Storage, Cloud and Digital Power. For Data Storage,
a total of 35 solutions are released.
Getting Started
Log in to the SCT website and click “Solution”-> “Product”-> “Data
Storage” -> “All-Flash Storage” -> “All-Flash SAN” -> “Design Solutions”.
Step 1: Enter Requirements Storage Configuration Guide
Set the parameters based on

user requirements — Basic
and Detailed requirements.
Basic Requirements ①apply
to rapid designs, and Detailed You can choose different service types with
Requirements ②apply to your special requirements
more professional designs.
Each application scenario has the

optimal parameter values by default.
You can modify the parameter values.
Click to view the help description
of a parameter
Click Export and then check

the materials you need to
download them.
Step 2: Configuration Selection (1) Storage Configuration Guide
eDesigner automatically recommends a product model

according to the requirements on the IOPS and capacity eDesigner automatically recommends device models based
that you have entered. on the requirements and capacity planning results. The device
models are displayed on the Product Configuration page. You
can modify the device models and related parts in a project to
achieve the best configurations.
You can flexibly set the disk

type, storage pool quantity,
RAID type, disk quantity, and
hot-spare policy.
All specifications of the recommended

device are displayed in this slide. You can
modify the parameter values.
Note:
The parameter values are displayed in red(such as 2) are
displayed on the parameters to indicate that the parameter
values are manually modified and are no longer the default
values. Click to restore the default setting.
You can flexibly set the port

type and port numbers.
eDesigner automatically calculates storage performance and

capacity based on the requirements. The calculation results are
displayed on the Product Configuration page. You can manually
adjust the capacities and scales based on the actual conditions so
that the solution meets your requirements.
eDesigner calculates
the IOPS/bandwidth
based on the
recommended
configuration.
eDesigner calculates the

capacities based on the
configuration.
All the software licenses that the

recommend product can support are
listed here, you can choose Yes or No
by special requirements
DME Storage is a
storage management
tool. You can configure
the license and service
years at here for this
tool.
Step 3: Service Design (1)
eDesigner also supports service design. On the Basic

Service page, you can choose Service Type, Service Level,
Service Duration and Storage Data&Device Retention
Service, and Technical Support Service based on your
needs.
Step 3: Service Design (2)
On the Professional Services page, you can choose

Traditional DR and Backup Service, Storage-Data
Migration Service and Planning, Design, and
Implementation Service based on your needs.
Step 4: Finish Design (1)
Storage Configuration Guide
After design, you can click

“Generate Device” to
Click Export and then check generate device configuration.
the materials you need to
download them.
Step 4: Finish Design (2) Storage Configuration Guide
On the Physical Specifications page, you can view information such as

Volume, Weight and Power Consumption.
Step 4: Finish Design (3) Storage Configuration Guide
On the Result Visualization page, you can graphically view information

such as Hot Spare Capacity, Effective Capacity, and Performance.
Contents

4. Obtaining Help
Placing Order (1)
After all errors are cleared, click in the Operation column to calculate again, because it depends on the
configuration result of OceanStor Dorado 5000 product.
There are 11 errors in this configuration, all

of which must be corrected and eliminated
before the order can be submitted.
Click the blue text under Config

Name Version to access the
corresponding product configuration
page and correct the errors.
Placing Order (2)
Change NO to Yes in the Setting

column to clear the error.
Move the cursor to the

error or alarm icon
to view the error or
alarm cause.
Placing Order (3)
For storage products, the following

parts must be configured: controller
enclosure, interface modules, disk
components and software License.
For non-quoted items, such as
power cables, the configuration
result is not displayed.
Placing Order (4)
After all product errors are cleared,

recalculate the configuration data of
dependent products and click
Submit in the lower right corner to
submit the quotation.
Sharing Solutions
You can quickly share your solution or quotation with other colleagues: My Quotation -> Draft/Submmitted, check the solution which you want
to share and then click Share.
If you have not submitted the solution, it is in the

Draft. If the solution is submitted successfully,
choose Submitted.
Note:
• Address Book: After sharing a solution,
the email address of the user who wants
the solution is automatically uploaded
to the Address Book. You can click the
email to share again.
Contents

4. Obtaining Help
Feedback and Help
On any page, ①Click Help to obtain help

resources such as user guides, permission guide,
training guide, and online support. ②Click Advice
to send your feedback to eDesigner.
To help eDesigner technical personnel reply in a
timely manner, leave your contact information
(email, phone number, or employee ID).
③Please send an email to:

UniSTAR SCT@huawei.com
intelligent world.



HCSA-Presales-Storage V4.0 Training Material

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

HCSA-Presales-Storage V4.0 Training Material

Uploaded by

Copyright:

Available Formats

Basic Knowledge of Data Storage

1. Positioning and Classification of Data Storage

2. Basic Concepts of Disks

3. Basic Concepts of Disk Redundancy Protection

4. Data Storage Development Trend

iSCSI or Fibre Channel

⚫ File storage is mainly used to store unstructured data.

⚫ Object storage, also known as object-based storage

Controller enclosure Controller enclosure

Scale up directly connected through backplane circuits,

Horizontal scaling: 3 nodes at least,

Node 1 Node 2 Node 3 Node N

1. Positioning and Classification of Data Storage

2. Basic Concepts of Disks

3. Basic Concepts of Disk Redundancy Protection

4. Data Storage Development Trend

Size Interface protocols

⚫ A hard disk drive (HDD) is a traditional hard disk. It consists of a platter,

⚫ A solid-state disk or solid-state drive (SSD) is also called an electronic

Item SSD HDD

MTBF (h) 1,200,000 1,600,000 1,200,000 2,000,000 2,000,000

An SSD is made up of a solid-state

1. Positioning and Classification of Data Storage

2. Basic Concepts of Disks

3. Basic Concepts of Disk Redundancy Protection

4. Data Storage Development Trend

Data arrangement Dedicated parity Distributed parity

Fault tolerance None High Medium Medium High High

Hot backup None Yes Yes Yes Yes Yes

Available capacity All 1/n (n-1)/n (n-1)/n (n-2)/n 1/m

File Original file

1. Positioning and Classification of Data Storage

2. Basic Concepts of Disks

3. Basic Concepts of Disk Redundancy Protection

4. Data Storage Development Trend

Online [The People's Bank of China

Latency 5 ms -> 0.5 ms 99.999% -> 99.9999% Reliability

Ultrasonic sensor Data collection and

Software and hardware

neutrality by neutrality by Carbon Carbon neutrality by billion by 2022.

2060 2050 neutrality by 2050

Copyright© 2023 Huawei Technologies Co., Ltd.

The information in this document may contain predictive

1. SAN Storage Overview

2. SAN Storage Networking

3. SAN Network Protocols

4. SAN Storage Technologies

5. SAN Storage Products and Application Scenarios

Server Server Server

JBOD JBOD JBOD

A SAN consists of a communication infrastructure, which

Storage device Storage device Storage device From: SNIA

Provide centralized management and monitoring tools, reducing O&M OPEX.

Centralized SAN Distributed SAN

Network NVMe over Fabric using RDMA;

4 Gbit/s, 8 Gbit/s, 40 Gbit/s, 56

Application server Application Application server

Protocol SCSI/FC/ATA FC/iSCSI TCP/IP

1. SAN Storage Overview

2. SAN Storage Networking

3. SAN Network Protocols

4. SAN Storage Technologies

5. SAN Storage Products and Application Scenarios

Fabric layer SAN switch SAN switch Cable