ISTORE1: Hardware Prototype

1
ISTORE A Scalable System Architecture for 2006
http//iram.cs.berkeley.edu/istore

• VIRAM System on a Chip
• Prototype scheduled for tape-out 1H 2000
• 0.18 um EDL process
• 16 MB DRAM, 8 banks
• MIPS Scalar core and caches _at_ 200 MHz
• 4 64-bit vector unit pipelines _at_ 200 MHz
• 4 100 MB parallel I/O lines
• 17x17 mm
• 2 Watts
• 25.6 GB/s
  memory bw

• ISTORE-1 Hardware Prototype
• Hardware plug-and-play intelligent devices with
  self-monitoring, diagnostics, and fault injection
  hardware
• intelligence used to collect and filter
  monitoring data
• diagnostics and fault injection enhance
  robustness
• Trial run of board fab complete 2-node system
  running
• 80-node system scheduled for 2Q 2000

Serial I/O

• Minimizes pins, power, and area.
• test chip completed 3Q 9
• link bandwidth of 1Gbit/sec
• lt 1.2 mm2 area
• single-poly, 4-metal, 0.35 um CMOS
• lt 160 mWatts, 3.3V supply

Intelligent Disk Brick Portable PC Processor
Pentium II DRAM Redundant NICs (4 100 Mb/s
links) Diagnostic Processor
Memory (8 MBytes)

• Intelligent Chassis
• 80 nodes, 8 per tray
• 2 levels of switches
• 20 100 Mb/s
• 2 1 Gb/s
• Environment Monitoring
• UPS, redundant PS,
• fans, heat and vibration sensors...

I/O
4 Vector Lanes
Memory Xbar
18 GB Disk
Memory (8 MBytes)
Half-height canister

• ISTORE in 2006
• IBM MicroDrive
• 1.7 x 1.4 x 0.2
• 9 GB, 50 MB/s (based on 340 MB, 5 MB/s in 99)
• ISTORE node
• MicroDrive IRAM
• Xbar switches growing by Moores Law
• 64 x 64 (based on 16x16 in 99)
• ISTORE rack (19 x 33 x 84)
• 1 tray (3 high) gt 16 x 32 gt 512 ISTORE nodes
• 20 trays switches UPS gt 10,240 ISTORE nodes

IRAM Applications
Intelligent PDAs speech, vision, camera,
radio, game, phone, remote
Peak GOPS
Sustained GOPS
of Peak
Image Composition
6.4
100.0
iDCT
6.4
30.7
Color Conversion
3.2
96.0
Image Convolution
3.2
98.7
Integer MV Multiply
3.2
86.5
Integer VM Multiply
3.2
93.7
FP MV Multiply
3.2
87.5
ISTORE Software Approach
3.2
FP VM Multiply
99.6
AME Software goals 1) Availability
continuous model, not just working/broken 2)
Maintainability 40 of failures due to admin
errors 3) Evolutionary growth scaling over
time gt heterogeneity Use of introspection to
meet these goals 1) reactive
self-maintenance dynamic reaction to exceptional
events through self-monitoring 2) proactive
self-maintenance continuous online self-testing
and self-analysis, including fault
injection Definition of new AME benchmarks 1)
because benchmarks shape the field 2) initial
results benchmarks for software RAID quantify
performance/reliability tradeoffs
FP 1D FFT
70.0
1.4
2.0
ISTORE Applications

• Storage-intensive, reliable services
• infrastructure for thin clients, e.g., PDAs
• web services
• databases
• Scalable memory-intensive computations for ISTORE
  in 2006
• simulations n-body, sparse system solvers
• DIS benchmarks 3D FFTs, MoM
• signal processing QR Decomposition