You are on page 1of 46

Ljubiša Crnadak

Solution Manager
Cloud Core Solution Sales, CEE&Nordic Region
Virtualization Cloudification

HW and SW Decoupled
Pooled Fully
hardware distributed
resources architecture

1. On-demand self-service,
Decoupled 2. Broad network access,
Full 3. Resource pooling,
automation 4. Rapid elasticity
5. measured service.

2
3
207.17.117.20

Internet

Load
Balance

Pool
Only show single external
216.34.94.17 IP address to remote node

4
207.17.117.20

Packet # 1
Src - 207.17.117.20
Dest – 216.34.94.17 Internet

Load
Balance

Pool
Only show single external
216.34.94.17 IP address to remote node

5
207.17.117.20

Packet # 1
Src - 207.17.117.20
Dest – 216.34.94.17 Internet

LB Translate to Load
internal IP address Balance

Pool
Only show single external
216.34.94.17 IP address to remote node

6
207.17.117.20

Packet # 1
Src - 207.17.117.20
Dest – 216.34.94.17 Internet

LB Translate to Load
internal IP address Balance

Packet # 1
Src – 207.17.117.20
Dest – 172.16.20.1

Pool
Only show single external
216.34.94.17 IP address to remote node

7
207.17.117.20

Packet # 1
Src - 207.17.117.20
Dest – 216.34.94.17 Internet

LB Translate to Load
internal IP address Balance

Packet # 1
Packet # 2 - return
Src – 207.17.117.20
Dest – 207.17.117.20
Dest – 172.16.20.1
Src – 172.16.20.1
Pool
Only show single external
216.34.94.17 IP address to remote node

8
207.17.117.20

Packet # 1
Src - 207.17.117.20
Dest – 216.34.94.17 Internet

LB Translate to Load LB Translate to


internal IP address Balance external IP address

Packet # 1
Packet # 2 - return
Src – 207.17.117.20
Dest – 207.17.117.20
Dest – 172.16.20.1
Src – 172.16.20.1
Pool
Only show single external
216.34.94.17 IP address to remote node

9
207.17.117.20

Packet # 1
Src - 207.17.117.20
Dest – 216.34.94.17 Internet Packet # 2 - return
Dest - 207.17.117.20
Src – 216.34.94.17

LB Translate to Load LB Translate to


internal IP address Balance external IP address

Packet # 1
Packet # 2 - return
Src – 207.17.117.20
Dest – 207.17.117.20
Dest – 172.16.20.1
Src – 172.16.20.1
Pool
Only show single external
216.34.94.17 IP address to remote node

10
Internet Internet

Load Load
Client requests are Balance Balance Administrator sets ratio for
distributed evenly distributing Client requests
3:2:1:1

1 2 3 4 9 1 5 7 2 6 3 4
5 6 7 8 10 8 12 14 9 13 10 11

11
12
•Scale out as traffic grows •Capacity of resource pool is automatically

•Scale in as traffic decreases re-allocated between PS and CS in earthquake

•Rapidly recycle and reuse resources


Scale in for data traffic

•Fully automated scaling Voice


Capacity Scale out for voice traffic Voice
Capacity
Data Data
Capacity Earthquake Capacity

Tourist Tourist
Peak Non-Tourist Peak
Peak

Thailand Case Natural Disaster Case

13
UE1 UE2

LB

Subscriber Subscriber
registers here register here

Active Standby Standby Active

VNF

NFVI

8
14
UE1 UE2

LB

Subscriber Subscriber
registers here register here

Active Standby Standby Active

VNF

NFVI

8
15
UE1 UE2

LB

Subscriber Calls here Subscriber


registers here register here

Active Standby Standby Active

VNF

NFVI

8
16
UE1 UE2

• Resource utilization is only 50% as


LB
standby node is idle before becoming
Subscriber Calls here Subscriber active.
registers here register here

Active Standby Standby Active

VNF

NFVI

8
17
UE1 UE2

• Resource utilization is only 50% as


LB
standby node is idle before becoming
Subscriber Calls here Subscriber active.
registers here register here
• Both active and standby fail because of
Active Standby COTS failure.
Standby Active

VNF

NFVI

8
18
UE1 UE2

• Resource utilization is only 50% as


LB
standby node is idle before becoming
Subscriber Calls here Call dropped Subscriber active.
registers here register here
• Both active and standby fail because of
Active Standby COTS failure.
Standby Active

VNF

NFVI

8
19
UE1 UE2

• Resource utilization is only 50% as


LB
standby node is idle before becoming
Subscriber Calls here Call dropped Subscriber active.
registers here register here
• Both active and standby fail because of
Active Standby COTS failure.
Standby Active

VNF

NFVI

8
20
Traffic cannot be re-balanced quickly after scale out
Load Load Load Load Load Load

90%---------- 90%---------- 90%---------- 90%----------

10%---------- 10%----------

Subscriber register here New Subscriber register here

Service Processing Service Processing Service Processing Service Processing Service Processing Service Processing
+ + + + + +
User data User data Sale Out User data User data User data User data
+ + + + + +
Session data Session data Session data Session data Session data Session data

LB LB

Only New Only New


Subscriber’s Traffic Subscriber’s Traffic
21
Traffic cannot be re-balanced quickly after scale out
Load Load Load Load Load Load

90%---------- 90%---------- 90%---------- 90%----------

10%---------- 10%----------

Subscriber register here New Subscriber register here

Service Processing Service Processing Service Processing Service Processing Service Processing Service Processing
+ + + + + +
User data User data Sale Out User data User data User data User data
+ + + + + +
Session data Session data Session data Session data Session data Session data

LB LB

Existing Subscribers Only New Only New


Subscriber’s Traffic Subscriber’s Traffic
22 Call Traffic Call Traffic
Traffic cannot be re-balanced quickly after scale out
Load Load Load Load Load Load

90%---------- 90%---------- 90%---------- 90%----------

10%---------- 10%----------

Subscriber register here New Subscriber register here

Service Processing Service Processing Service Processing Service Processing Service Processing Service Processing
+ + + + + +
User data User data Sale Out User data User data User data User data
+ + + + + +
Session data Session data Session data Session data Session data Session data

LB LB

Existing Subscribers Only New Only New


Subscriber’s Traffic Subscriber’s Traffic
23 Call Traffic Call Traffic
Subscriber register here
Service Processing Service Processing Service Processing Service Processing Service Processing Service Processing
+ + + + + +
User data User data User data User data User data User data
+ + + + Scale In + +
Session data Session data Session data Session data Session data Session data

① ② ③ ④ ① ②

LB LB
LB LB

Call Traffic

24
Subscriber register here
Service Processing Service Processing Service Processing Service Processing Service Processing Service Processing
+ + + + + +
User data User data User data User data User data User data
+ + + + Scale In + +
Session data Session data Session data Session data Session data Session data

① ② ③ ④ ① ②

LB LB
LB LB

Call Traffic
Call Traffic

25
Subscriber register here
Service Processing Service Processing Service Processing Service Processing Service Processing Service Processing
+ + + + + +
User data User data User data User data
Scale In
User data User data Lost Call
+ + + + + +
Session data Session data Session data Session data Session data Session data

① ② ③ ④ ① ② There is no call
data

LB LB
LB LB

Call Traffic
Call Traffic

26
Traditional
N-Way
(All-Active)
Active-Standby

010101010001101010101001
Active 010101010101010101010001 Standby

Cloudified

27
VNF with 3-Tiers Cloudified Architecture
Distributed Session Database Layer

VM VM VM

Stateless Service Process Unit Layer

Orchestration
CloudIMS CloudEPC

Distributed Load Balancer Layer

VM VM VM

Cloud OS

Cloud Infrastructure

28
DPU
① DPU

SCU1 SCU2 SCU3 SCU4 SCU1 SCU2 SCU3 SCU4


④ Call
processing
④ processing
Call

② ③ ③ ②
Session Session
data data

User User User User


data data data Data
RDB RDB RDB RDB

Call traffic can be switched over to healthy VM quickly (~second) and no call loss.
29
Load Load Load Load Load Load Load Load Load

90%---------- 90%---------- 90%---------- 90%----------


72%---------- 72%---------- 72%---------- 72%---------- 72%----------

DPU

DPU

SCU1 SCU2 SCU3 SCU4 Scale Out SCU1 SCU2 SCU3 SCU4 SCU5
④ Call
processing



Session
data

User User User User


data data data data
RDB RDB RDB RDB

30
After scaling out, existing call traffic is re-balanced very quickly (~seconds)

DPU DPU

SCU1 SCU2 SCU3 SCU4 SCU5 SCU1 SCU2 SCU3 SCU4 SCU5
Scale In
Call

④ processing


Session Session
data data

User User User User


data data data data

RDB RDB RDB RDB

31
Flow Control Objective:
•Avoid system collapse because of traffic storm or unusual traffic.

• In NFV, CPU occupation information is If traffic control is only based on CPU load:
not accurate.
• Not Accurate: traffic is not balanced among VMs.

• Overestimate: Cause false flow control


Real Physical CPU Load from
CPU Load VM perspective
• Underestimate: Do not control traffic after overload. Cannot
60% handle signal storm in Telecom.
overestimate
V V V V
underestimate M M M M
40% 1 2 3 4

Traffic
Distribution

32
Host OS Occupation Idle VM Occupation
Enhance HA via COTS Agnostic Overload Control Mechanism:
Queue Flow Control
CPU, Memory, I/O,
Flow Control Message Delay
Notification

... Priority 1 Priority 1

...
Priority 2 Priority 2
... ...

Tag priority Priority N Message Message Execute Tag priority Priority N Message Message Execute
and put Dispatcher processor Flow and put Dispatcher processor Flow
messages to Control message to Control
different queue different queue VM (SCU/CCU/…)
VM (BSU/BSG/…)

• Under 1.5 times overload, ITU-T G.543 require that access success rate should reach 90%.
Non-Cloudified architecture cannot meet this requirement.
• After 2.8 times overload, Huawei IMS access success rate still reaches 90%.

33
• More than 50% network fault due to partially unhealthy behavior
(delay, jitter, packet loss, packet corrupted, etc.)
• Harder fault detection and recovery in layered cloud system

KPI-based Partially Unhealthy Detection and Auto-Healing


When Gap beyond expectation
Unhealthy VNF define system unhealthy Health VNF
VM KPI Monitor VM
History KPI Predetermined
Action
VM VM VM Gap VM VM VM
Success call ratio VM isolated
Registration ratio Real-time KPI VM migration
… …
VM VM VM VM

Supporting VNF/NS level fault self-healing

34
EMS:CloudU2000
CCE
alarm Real time monitoring of service Alarm occurs when
Fault Mgmt analyze Collect
KPI on VNF and VM KPI is going down

Collect service Action for


KPI Recovery

VNF1 VNF2

VM1 VM2 VM3 VM1 VM2 VM3

NFVI

35
CSC VNF1 VNF Pool

VNFC

KPI Collection &


LB VNFC Recovery Center

SBC1
VNFC
VNFC Pool

SBC2
CSC VNFn
… KPI Collection &
Recovery Center
VNFM

… VNFC
U2000/CCE

SBCn
KPI Collection &
LB VNFC Recovery Center

KPI Collection …

VNFC
Restful API
VNFC Healing
VNFC Pool
(Support 3rd
VNF)
VNF Healing
Sub-health
NFVI VIM
36
KPI based resource scaling
Real-time KPI
CAPS
Scale
"Zero" session
NO. of subs
vAPPs Monitoring
Call Success Rate
in / out interruption
Redundancy
Layered Synergy Active/standby Stateless N-Way VNF Cross-DC 99.999%
HA Mechanism New VM3 Reliability
Cloud VM3
VM2 Recreation VM2
VM1 VM1
OS Cluster N+M redundancy
Resources always
Cluster Hardware resource redundancy
HW available

Top N Failure and Deal Ways


Application Network
Single Single hard DC Cloud Guest OS
Abnormal Qos
server failure disk failure failure OS overhead Hang/Crash
terminated descending
Others Swap traffic to Others takes Read/write Others take
takes over another virtual over the failed Data from Cross-DC over the
App Layer redundancy
NONE
the failed one Network plane one’s traffic redundancy disk failed one

Re-Spawn VM VM migration
Cloud OS NONE NONE Re-Spawn VM NONE
(normal DC) to other DC
Reset VM

Hardware NONE NONE Isolated Isolated Manual check NONE NONE

37
Non-Cloudfication Cloudification

• 1+1 Redundancy, Non-distributed architecture • All-Active, Distributed architecture


Fully
distributed
• Data and service are coupled architecture • Data and service are de-coupled

• Centralized Database • De-centralized Database

• Resource reservation for Redundancy Pooled • Resource On-demand


hardware
• Low resource utilization efficiency resources • Maximize resource utilization

• Don’t rebalance after scale out • Auto-Scale with no any Call Lose
Full

• Auto-Scale in with Call Lose


automation
• Fault automatic detection and diagnosis,
auto-healing based on service KPI

38
TARGET ARCHITECTURE OF CLOUDEPC
Current Future

DB DB

PCRF CG
PCRF CG vCG
HLR-FE HSS-FE HLR-FE HSS-FE

Gr Gr R7 Gx R8 Gx
Gr S6a R7 Gx R8 Gx Ga

GGSN SGW/PGW GGSN + SGW/PGW vUGW


Gn
S11 S1-U Gn/S11
S1-U
Gn
SGSN MME SGGN + MME vUSN

S1-MME S1-MME
CloudEPC
BSC RNC eNodeB BSC RNC eNodeB

2G/3G PS 4G Unified
PS PS
Huawei CloudEPC VNF Architecture
OSS/BSS NFVO

EMS vU2000 vPRS VNFM


CLoudOpera
VNF CSM
vUSN vUGW vCG CloudEPC

NFVI
Virtualisation Layer
Fusionsphere (Openstack/KVM)
VIM
Fusion
COTS manager
+
E9000 S5500 CE6800 eSight
VNF Deployment Approach
Command Way Openstack As a Tool Huawei Openstack+MANO
(Manually, Low Efficiency) ( Semi-Automatic NFV ) ( Batch,Automatic,Dynamic NFV )

External Internal
External
Network1 Network1
External Internal
Network
Network2 Network2

OMU IPU SPU


SPU SPU SPU

OMU IPU SPU


VM VM VM
#setup network
Ifconfig eth0 0.0.0.0
brctl addbr br0
brctl addif eth0
#setup storage
qemu-img create -f qcow2 /vdisk.img 10G
#Create VM
virt-install \
-n mylinux \
--description " VM with RHEL 6" \
--os-type=Linux \
--os-variant=rhel6 \
--ram=2048 \
--vcpus=2 \
#Upload VNF
--disk path=/vdisk.img \
Huawei NFV Architecture: Comply with NFV and Keep Openness
NFV Management and
Orchestration
MANO (VIM + NFVO + VNFM )
Os-Ma NFVO
OSS/BSS Orchestrator
CloudOpera
Se-Ma Orchestrator
NFVO
Service, VNF and Infrastructure
Description
Or-Vnfm
• NFVO manage network service lifecycle
EMS 1
vU2000
EMS 2 EMS 3 Ve-Vnfm VNFM • (EPC, IMS, VOLTE…)
iManager U2000 VNF • NSD: NS descriptor, a set of files that describes the
CloudOpera
VNF vIMS vEPC
Manager(s)
CSM VNF topology and resource requirements during the
vGi-LAN
VNF 1
CloudIMS
VNF 2
CloudEdge
VNF 3
CloudMSE
Or-Vi lifecycle of the NS
Vn-Nf Vi-Vnfm
NFVI
Virtual Virtual Virtual
VNFM ( S-VNFM / G-VNFM)
HuaweiStorage
Computing Cloud OSNetwork VIM • VNFM manage VNF lifecycle
FusionSphere/3 rd Party Nf-Vi Fusion
Virtualised • (vUSN, vUGW, vCSCF…)
NFVI Virtualisation Layer
Vl-Ha
Infrastructure
Manager
Manager(s) VNFD: VNF descriptor, a set of files that describes
Huawei Hardware resources
COTS +eSight the VM topology and resource requirements during
Computing Storage Network
rd
HardwareE9000/3 Party
Hardware Hardware the ifecycle of the VNF

Execution reference points Other reference points Main NFV reference points
VIM
MANO • Manage physical resources lifecycle
Create VM , delete VM
Collect hardware alarm info.
CloudEPC On Board with MANO
Each VM Flavor
e.g. to deploy CloudEPC
Upload of NSD/ Directory Network
1
Based on TOSCA NFV Orchestrator Policy
•Network Service Descriptor Written in XML / yaml CloudEPC

2 Upload batch of the VNFD, Call VNFM


vUSN/vUGW/vCG
NS Descriptor
VNF Manager Read VNFD
Service ID (CloudEPC) VNF Manager
3 Check/Reserve the
Vendor (Huawei)
vUSN Description Resource
Version (VxRx)
5VMs for 5VMs for 3VMs for
VNFD Index vUGW Description vUSN vUGW vCG
4 Call VIM to Create
VNFFGD
…. /Delete the VM/
VLD
VIM VIM VIM Download IMG
Scale Policy VNF1 VNF2 VNF3

COTS
MWC 2017 – NFV Solution: Best Technology Enabler
Thank you
Copyright©2017 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.

You might also like