Professional Documents
Culture Documents
Cisco InfiniBand - Today and Tomorrow PDF
Cisco InfiniBand - Today and Tomorrow PDF
Jamie Riotto
Sr. Director of Engineering
Cisco Systems (formerly Topspin Communications)
jriotto@cisco.com
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 1
Agenda
• InfiniBand Today
– State of the market
– Cisco and InfiniBand
– InfiniBand products available now
– Open source initiatives
• InfiniBand Tomorrow
– Scaling InfiniBand
– Future Issues
• Q&A
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 2
InfiniBand Maturity Milestones
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 3
Cisco Adopts InfiniBand
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 4
Cisco and InfiniBand
The Server Fabric Switch
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 5
Cisco HPC Case Studies
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 6
Real Deployments Today: Wall Street Bank
with 512 Node Grid
Fibre Channel and GigE
Existing SAN LAN
connectivity built
Networks seamlessly into the cluster
GRID
I/O 2 TS-360 w/ Ethernet and Fibre
Channel Gateways
2 96-port
Core TS-270
Fabric
Edge 23 24-port
Fabric TS-120
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 7
NCSA
National Center for Supercomputing Applications
174 uplink
cables
Edge 29 24-port
Fabric TS120
512 1m
cables
520 Dual CPU Nodes
18 Compute 18 Compute 1,040 CPUs
Nodes Nodes
Fault 12 96-port
Tolerant TS-270
Core
Fabric 1,068 5m/7m/10m/15m
uplink cables
Edge 89 24-port
Fabric TS-120
1,066 1m
cables
12 Compute 12 Compute
Nodes Nodes
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 9
Large Government Lab
Worlds Largest Commodity Server Cluster – 4096 nodes
• Application:
High Performance Super 8x SFS TS740
288 ports each
Computing Cluster
Core
• Environment: Fabric
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 10
InfiniBand Products Available
Today
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 11
InfiniBand Switches and HCAs
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 12
Integrated InfiniBand for Blade Servers
Create “wire-once” fabric
HCA
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 13
Ethernet and Fibre Channel Gateways
Unified “wire-once” fabric
Server Cluster
SAN LAN/WAN
Server Fabric
Fibre Channel to InfiniBand gateway for storage Ethernet to InfiniBand gateway for LAN
access access
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 14
InfiniBand Price / Performance
InfiniBand
10GigE GigE Myrinet D Myrinet E
PCI-Express
Data Bandwidth 950MB/s 900MB/s 100MB/s 245MB/s 495MB/s
(Large Messages)
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 15
InfiniBand Cabling
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 16
Host Drivers for Standard Protocols
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 17
OS Support
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 18
The InfiniBand Driver Architecture
APPLICATION
INFINIBAND SAN
NETWORK
SAN
LAN/WAN SERVER FABRIC
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 19
Open Software Initiatives
• OpenIB.org
– Topspin primary authors of major portions
including IPoIB, SDP, SRP and TS-API. Cisco will
continue to invest.
– Current protocol development nearing production
quality code. Expect release by end of year.
– Charter has been expanded to include Windows
and iWarp
– MPI will be available in the near future (MVAPICH
0.96)
• OpenSM
• OpenMPI
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 20
InfiniBand Tomorrow
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 21
Looking into the future
• Cost
• Speed
• Distance Limitations
• Cable Management
• Scalability
• IB and Ethernet
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 22
Speed: InfiniBand DDR / QDR, 4X / 12X
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 24
IB Distance Scaling
• IB Short Haul
– New Copper drivers
– 25 – 50 Meters (KeyEye)
– 75 - 100 Meters (IEEE 10Ge)
• IB Wan
– Same Subnet over distance (300 KM target)
– Buffer / Credit / Timeout issues
– Applications: Disaster Recover, Data Mirroring
• IB Long Haul
– IB over IP (over SONET?)
– utilizes existing public plant (WDM, Debugging, etc)
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 25
Scaling InfiniBand
• Subnet Management
• Host-side Drivers
MPI
IPoIB
SRP
• Memory Utilization
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 26
IB Subnet Manager
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 27
Subnet Management Challenges
• Cluster Cold Start times
–Template Routing
– Persistent Routing
• Cluster Topology Change Management
– Intentional Change - Maintenance
– Unintentional Change – Dealing with Faults
• How to impact minimum number of connections
• Predetermine fault reaction strategy?
• Topology Diagnostic Tools
– Link/Route Verification
– Built-in BERT testing
• Partition Management
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 28
Multiple Routing Models
• Minimum Latency Routing:
– Load-Balanced Shortest-Path Routing
• Minimum Contention Routing:
– Lowest-Interference Divergent-Path Routing
• Template Driven Routing:
– Supports Pre-Determined Routing Topology
– For example: Clos Routing, Matrix Row/Column, etc
– Automatic Cabling Verification for Large Installations
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 29
IB Routing Challenges
• Static / Dynamic Routing
– IB impliments Static Routing through Linear Forwarding
Tables at each chip
– Multi-LID Routing enables Dynamic Routing
• Credit Loops
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 30
Multi-LID Source-Based Routing Support
1,2,3,4
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 31
New IB Peripherals
• CPUs?
• Storage
– SAN
– NFS-RDMA
• Memory (coherent / non-coherent)
• Purpose built Processors?
– Floating Point Processors
– Graphics Processors
– Pattern Matching Hardware
– XML Processor
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 32
THANK YOU!
Session Number
Presentation_ID © 2005 Cisco Systems, Inc. All rights reserved. Cisco Public 33