Unified Wire Felix Marti, felixchelsio'com

1
ChelsioCommunications
Unified WireFelix Marti, felix_at_chelsio.com Open
Fabrics Alliance Workshop Sonoma, April 2008
2
T3 Single Chip Can Do All

• A True Unified Wire
• Ideal for clustering
• Runs IB applications using iWARP
• Ideal for storage
• Runs FC applications using iSCSI
• Does not require forklift upgrade of the switch
  infrastructure
• Ideal for Networking
• Fully compatible with VMWare and Xen
• TOE provides predictable bandwidth and latency
• Integrated traffic management for seamless
  sharing of infrastructure
• Integrated classification and firewall support
• Rich roadmap
• On the motherboard in 2009 do clustering and
  storage for zero incremental cost and no
  performance penalty
• Software remains the same throughout leverage
  SW investment

Delivers the ROI of Fabric Convergence Addresses
InfiniBand FibreChannel Networking markets
2
3
Robust, Tested, Seasoned

• iWARP
• Sun Lustre team using T3 iWARP
• UCLA running T3 iWarp across the campus and to
  Reno _at_SC07
• 128-node Sandia cluster in use for 6 months with
  T3 iWARP
• Several Tier-1 server OEM wins
• TOE
• TOE shipping for 2 years
• Osaka University operating a 700 node TOE cluster
  for 1 year
• In the top 50 super computers in the world
• iSCSI
• Several Tier-1 storage OEM wins
• NIC
• Several Web 2.0 OEM wins (highest IOPs rate NIC)
• Holder of all current land speed records
• Tremendous virtualization facilities

The Enterprise Grade Unified Wire Solution
3
4
Sampling of Cards
4
5
Highest Performance

• Highest performance VMware implementation
• 10Gb on Receive and Transmit (1 VM on Tx, 2 VMs
  on Rx)
• Highest performance iSCSI implementation
• 900K IOPs, 2200 MB
• More than twice the performance of 4G FC for
  Oracle Exchange
• Highest performance NIC adapter
• Holds all the IPv4 and IPv6 records
• 16M packets/sec ideally suited for Web 2.0
  applications
• Highest transaction rate NIC on the market
• Highest performance TOE adapter
• 10GbE at lt 5 CPU with standard frames for 1 to
  many K connections
• Ability to accelerate the legacy sockets
  applications
• Fast error recovery delivers consistent and
  scalable performance
• Only 10Gb TOE with a WHQL-ready Chimney driver,
  and in FBSD kernel
• High performance clustering solution
• 6usec MPI latency, RDMA Read RTT/2 lt5usecs

Highest performance solution across all
dimensions Highest perforamance NIC-only, TOE,
iSCSI, RDMA
5
6
Performance Benchmark
OSU MPI Benchmark
iSCSI over Linux IOPS
iSCSI over Linux Throughput (MB/s)
7
Performance Benchmarks (2)
6.56 usecs
8
Thank You!