Intro Presentation - Globalpress Connection

Download Report

Transcript Intro Presentation - Globalpress Connection

Chelsio
Communications
Ethernet Unified Wire
November 15, 2006
Kianoosh Naghshineh, CEO
[email protected]
www.chelsio.com
Pioneering Unified Wire
LAN
SAN
Unified 10Gb Ethernet Network
HPC
SAN
LAN
HPC
NAS
NAS
Convergence Benefits
Lower Total Cost of Ownership
• Improves CPU efficiency
• Minimizes software licenses
• Simplifies data center wiring
• Leverages staffing skills & tools
Higher Performance and New Apps
• Improves cluster performance
• Lowers application latency
• Faster backup and recovery
• Enables storage applications
Simplified network architecture – reduced operating costs
CHELSIO CONFIDENTIAL
2
Key Market Drivers
Criteria
Unit Growth
Market Drivers / Enablers
 3x volume growth in the 10GbE NIC market in 2006 (synergy report)
 3x volume growth in 10GbE switch market in 2006 (Dell’Oro)
 iSCSI market growing at 50% per quarter
 High density 10GbE switches available now
 OpenFabrics released  can run IB apps unmodified, channel product
Infrastructure  Chimney OS released  All NICs become TNICs in channel
 iSCSI Target & Initiator ready  can begin to replace FC
 Middleware partnerships in place  can run FC apps unmodified
Prices
Standards
CHELSIO CONFIDENTIAL
 XFP over 12 months: prices halved
 CX4 switch port at $700/port list price now
 10G CX4 RNIC at $995. Expected to be halved by 4Q07
 10G CX4 (copper media) introduced and shipping
 10G-baseT switches and cards expected by 1Q07
3
Current Shipping Products
CX4 copper or
XFP SR/LR optics
PCI-Express or
PCI-X bus interface
T210 10GbE
Protocol Engine
N210 10GbE
Server Adapter
T204 4GbE
Protocol Engine
Full TCP/IP Offload (TOE)
Full iSCSI Target solution
CX4 copper or XFP fiber
10Gb Ethernet NIC
Linux, Windows, Solaris
Turn-key software drivers
4-port Gigabit Ethernet
Full TCP/IP Offload (TOE)
Full iSCSI Target solution
Highest performance and
CPU efficiency
Easy-to-use server adapter
for 10GbE connectivity
Best price/performance
for servers & IP SANs




2nd generation products shipping in volume since Feb 2005
Over 140 customers in 20 countries using Chelsio adapters
Strong foundation for 3rd generation ‘unified wire’ solution
Holder of all but one of the current Land Speed Records
CHELSIO CONFIDENTIAL
4
Introducing Terminator 3 (T3)
PCI-Express or
PCI-X bus interface
T3 ASIC








PCIe x8 or PCI-X 2.0
2x10G or 2x1G Ethernet
NIC+TOE+iSCSI+RDMA
Up to 1M connections
ACPI, IPMI, SMBus, I2C
0.13um TSMC LV
~11W typical power
37mm FCGBA
CHELSIO CONFIDENTIAL
T302 2x1GbE
PCI-Express
T310 10GbE
PCI-Express or PCI-X 2.0






PCIe x8 or PCI-X 2.0
10GbE CX4 or XFP
NIC+TOE+iSCSI+RDMA
32K connections
iSCSI Target+Initiator
Full-duplex 10Gbbps
line-rate performance
 ½ length form factor







PCIe x4
2-port Gigabit Ethernet
NIC+TOE+iSCSI+RDMA
32K connections
iSCSI Target+Initiator
2 x 1000BASE-T ports
Low-profile form factor
5
Superior Performance
Chelsio T110 TOE versus 10GbE NIC Network Performance
Higher Throughput (Gb/s)
Lower CPU Utilization (%)
T110
N110
Intel
S2io
8
7
6
5
4
3
2
1
0
120%
100%
80%
60%
40%
20%
0%
TX
RX
TX
RX
Source: Independently verified by VeriTest, Inc.
Test tool: netperf
Test configuration: 2 systems connected through 10GbE switch running single TCP channel with 1500-byte Ethernet frames
System configuration: AMD Opteron 248 2.2GHz uniprocessor running Linux kernel 2.6.6





Performed May 2003 by the world’s leading independent lab, VeriTest, Inc.
TOE delivers ~2X network throughput vs. 10GbE NICs
TOE shows ~1/2X CPU utilization vs. 10GbE NICs
Net-net: TOE shows ~4X network efficiency vs. NICs
No Jumbos, one connection to thousands of connections
CHELSIO CONFIDENTIAL
6
Independent HPC Benchmarks
Validate Performance
Bandwidth vs Message Size
10GigE
7000
SDP/IBA (Event)
10GbE has higher
bandwidth than
Infiniband & Myrinet
Bandwidth (Mbps)
6000
SDP/IBA (Poll)
5000
SDP/Myrinet (Event)
4000
SDP/Myrinet (Poll)
3000
2000
1000
25
6K
64
K
16
K
4K
1K
25
6
64
16
4
1
0
Message Size (bytes)
10GbE has lower
latency than
Infiniband
Latency (usec)
Latency vs Message Size
10GigE TOE
100
90
80
70
60
50
40
30
20
10
0
SDP/IBA (Event)
SDP/IBA (Poll)
SDP/Myrinet (Event)
SDP/Myrinet (Poll)
1
2
4
8
16
32
64 128 256 512 1K
2K
4K
8K 16K
Message Size (bytes)
CHELSIO CONFIDENTIAL
Testing performed by Los Alamos National Laboratory and The Ohio State University
7
RDMA & iSCSI







12
PCI-X 133MHZ
1.8GHZ Dual CPU OPTERON
1GB Memory
Mellanox 4X SDR IB PCI-X HCA
Chelsio T320X 10GbE PCI-X RNIC
Point to point networks
Using Open-IB derivative
benchmarks
10
8
6
Chelsio
4
Mellanox
2
0
1B
1B
Read
Write
(usec) (usec)
700
100
90
600
80
CHELSIO CONFIDENTIAL
50
300
40
% CPU
60
400
30
200
20
100
10
IO Size
256K
64K
0
16K
0
4K
20 Initiators – MS iSCSI 2.0, 1GbE
Dell 2950, 2 x Intel 3GHz EM64T Dual Core Woodcrest,
8GB DDR-II 667MHz memory
Linux 2.6.17 SMP x86_64.
T320e (T3b).
iSCSI Target 2.0 release.
Ramdisk. No CRC, TOE mode
70
512
IOPs x 1000
500
Read
Write
%CPU Rd
%CPU Wr
8
T3 Feature Summary
Feature
Description
Hardware
 Integrated native PCIe 8x and PCI-X 2.0 266MHz interface
 Integrated 2 GigE or 2 10GbE ports with CX4/KX4 compliant XAUI
Architecture
 Cut-through processing architecture delivers 10Gbps wire rate & ultra low latency
 Scalable to 100Gbps and extendable to additional protocols (IPSec, SSL, XML, NFS)
 Flexible in cost, power, performance, programmability
Management
 PXE, EFI, Power Management
QoS
 Multiple queues for bandwidth resource provisioning and allocation
 Multiple channels for simultaneous high bandwidth/low latency operation
Virtualization
 Extensive receive and transmit features to support virtualization
Filtering
 Tens of thousands of filter rules in hardware, configurable in software, used for firewalls and virtualization
Traffic Manager
 Simultaneous per-connection and per-class rate control and bandwidth allocation
TOE
 4x performance vs NIC
 Implements both Full and Partial TCP offload
 High performance even in demanding network environments (large RTT, packet loss)
DDP
 Zero-copy steering of payload data directly into application space for extremely low CPU utilization
 No modification to applications
 Line rate receive on 1 connection
iSCSI
 16x performance vs NIC
 Chelsio offers complete commercial-grade iSCSI software stack for Linux for FREE
iWARP RDMA
 RDMAC & IETF compliant
 RNIC-PI, kDAPL and OpenIB software interfaces
CHELSIO CONFIDENTIAL
9
Competitive Analysis

Chelsio competes primarily with companies that develop Ethernet
adapter cards and/or integrated circuits
 Indirect competition from other fabric technologies such as InfiniBand and Fibre
Channel

Chelsio delivers a best-in-class, full-featured and seasoned solution
NIC
TOE
iSCSI
RDMA
10GbE
GbE
2-port
PCI-2X
PCIe
VLIW
Chelsio
►
►
►
►
►
►
►
►
►
►
Broadcom
►
►
►
►
►
Emulex/Silverback
►
►
QLogic
►
►
Intel
►
►
Neterion
►
►
NetXen
►
CHELSIO CONFIDENTIAL
►
►
►
►
►
►
►
►
10
Why Chelsio








Very High Performance
 Very Low Latency
 High Transaction Rate
 High Bandwidth
Correct architecture
 Data Flow Processor produces a Scalable and Ideal Performance Profile
Value added features
 Integrated Traffic Manager
 Filtering
 Virtualization
3rd Generation – Robust, Seasoned
 Maximum guarantee of no late arriving QA bugs
Integrated High Performance ULP solutions – run IB or FC apps unmodified
 Commercial grade iSCSI stack integrated with hardware DDP facilities
 Fully integrated OpenFabrics RDMA software stack
Broad product offering – future protection
 2x1Gb or 2x10Gb, iSCSI, RDMA, TOE, Chimney, NIC
 PCI-Express, PC-X 2.0, PCI-X 1.0
Reliable, predictable supplier (ISO 9001 certified)
 The only vendor with experience shipping Offload products to enterprise OEMs
Availability of software-only products
CHELSIO CONFIDENTIAL
11
Thank You
CHELSIO CONFIDENTIAL
12