1. Emerging Storage Options for Server Blade Architectures Server Blade Summit 2005

2. 2 Storage Options for Server Blade Architectures Today’s blade server landscape Today’s blade server landscape Where are we going next? Where are we going next? New blade server architectural challenges and their impact in the data center New blade server architectural challenges and their impact in the data center Blade server storage choices and their key value propositions Blade server storage choices and their key value propositions

3. 3 Blade Server Complexity Back Plane Support Infiniband Fibre Channel CPU(s) 2 Local Mirrored Discs Chipset Memory Ethernet E I F S SAS Design Complexity 2 separate backplane fabrics Multiple “I/O down” ICs per blade 2 drives per blade (mirrored pair)

4. 4 Storage Connectivity for Server Blades Back Plane Support Infiniband Fibre Channel Ethernet CPU(s) Chipset Memory E I F S SAS DAS SAN NAS LAN TYPES OF GATEWAY SAS iSCSI TOE iSCSI FC SCSI SAS SATA Storage Connect Gateways are used for near line storage as well as fabric storage FC and iSCSI are SAN connect for block level transfers SAS and iSCSI is used for near cabinet storage NAS is used for ease-of-use and access to storage as the file transfer model

5. 5 Blade Chassis Storage Back Plane Support Infiniband Fibre Channel CPU(s) Chipset Memory Ethernet E I F S SAS DAS SAN NAS LAN TYPES OF GATEWAY SAS iSCSI TOE iSCSI FC SCSI SAS SATA Blade Chassis Storage “Semi-local” storage shared by a small number of blades It is not truly “SAN” because it is not available to all servers on the Fabric It is not truly “DAS” because it is not exclusively owned by a single server It is a new “intermediate” level in the storage hierarchy It is a compromise between the need to eliminate per-blade storage, yet maintain adequate performance for the virtual memory and high-use cached objects Architectural variances exist

6. 6 Diskless Server Blade Architecture Server blades are diskless, stateless and used as pool of generic assignable “workers” Server blades are diskless, stateless and used as pool of generic assignable “workers” Simple “size” parameters such as compute power (CPU), memory capacity, aggregate I/O bandwidth Simple “size” parameters such as compute power (CPU), memory capacity, aggregate I/O bandwidth A Global Resource Manager can assign and select desire boot image, applications or selective links to appropriate data volumes A Global Resource Manager can assign and select desire boot image, applications or selective links to appropriate data volumes Chassis Storage provides low latency access to storage for the server blades Chassis Storage provides low latency access to storage for the server blades It is a compromise between per-blade drives and SAN-resident drives It is a compromise between per-blade drives and SAN-resident drives It is especially critical for per-blade virtual memory (page swap) It is especially critical for per-blade virtual memory (page swap)

7. 7 Impact on Data Center Architecture server chassis leaf switch leaf switch server chassis leaf switch leaf switch Racks of server blades RAID Hard storage assets RAID Other Storage Elements tape, routers, etc tape, routers, etc. Storage Management & Storage Virtualization Layer Multi-tier network Infrastructure Cluster of intelligent Gateways SAN or appliances Blade Chassis Storage

8. 8 SAS is an ideal storage interconnect for Blade Server Chassis Low latency to disks is critical for many applications (Transactional) Low latency to disks is critical for many applications (Transactional) Multiple servers running different applications and different OS, will place highly random load demands on disks Multiple servers running different applications and different OS, will place highly random load demands on disks Transactional performance is critical where multiple servers collectively present diverse load Transactional performance is critical where multiple servers collectively present diverse load Advantage in Price/Performance Advantage in Price/Performance Storage amortized over multiple blades provides better cost structure Storage amortized over multiple blades provides better cost structure Reliability (MTBF) is critical Reliability (MTBF) is critical Multiple servers dependent on the storage for boot and application - therefore reliability is critical Multiple servers dependent on the storage for boot and application - therefore reliability is critical Storage disks mechanically will work hard, therefore robustness is critical Storage disks mechanically will work hard, therefore robustness is critical Ability to mix and match based on storage needs Ability to mix and match based on storage needs SAS – SAS or SAS – SATA SAS – SAS or SAS – SATA

9. 9 SAS/SATA IC or HBA End user ability to match drives to balance storage to the job and trade-off performance, capacity, reliability, and cost End user ability to match drives to balance storage to the job and trade-off performance, capacity, reliability, and cost Server vendor can create multiple market segments for the server blades Server vendor can create multiple market segments for the server blades I/O Vendor will benefit from having single chip or adapter serving multiple segments of server blade market I/O Vendor will benefit from having single chip or adapter serving multiple segments of server blade market Server blades for the midrange market will be first to adopt SAS as the replacement to U320 Server blades for the midrange market will be first to adopt SAS as the replacement to U320 SAS/SATA Value Proposition server chassis leafswitch leafswitch SAS/SATA Usage Model High Performance SAS Drives for Data Reliable SAS Drives for Mirrored Boot Pair High Capacity SATA Drives for Backup

10. 10 SAS and SATA Drives Enterprise vs. Desktop Applications Power management Power management Greatest device flexibility Greatest device flexibility  Simplest configuration setup Simplest configuration setup Highest reliability and availability Highest reliability and availability First to volume disk interconnect First to volume disk interconnect Greater scalability Greater scalability Most cost effective for reference & sequential-type data Most cost effective for reference & sequential-type data Best performance for transactional data Best performance for transactional data Lowest cost per gigabyte Lowest cost per gigabyte Highest performance capability SAS (SCSI, FC) SATA (ATA)

11. 11 Desktop vs. Enterprise Drive Parameters Enterprise Class Desktop Class Performance Variable Sector Size Indicators Rotational Vibration Reliability Dual Port Data Integrity 13msec@ 7200 NoNone 5-12 radians/sec/sec 600K hours @ 8x5 duty NoNo 5.7msec@ 15K YesLED >21 radians/sec/sec 1.2M hours @ 24x7 duty YesYes

12. 12 Contact Information Ken Zarrabi Strategic Marketing & Business Development RAID Storage Adapters, LSI Logic Tel : 408-433-4527 Fax: 408-433-4380 Email: kzarrabi@lsil.com

13. 13 Biography Ken Zarrabi joined LSI Logic in 2002 as a senior product marketing manager for the RAID Storage Adapters division. Mr. Zarrabi has more than 18 years of marketing, engineering and management experience in the high-tech industry. Prior to joining LSI Logic, he created and implemented marketing and business development plans for Adaptec and zDrive. His career experience also includes positions as a director of customer and product engineering at Fujitsu and six years as manager of customer engineering and program management at Quantum. Mr. Zarrabi earned a bachelor's degree in mechanical engineering with a minor in computer science from the University of Colorado.