Arbitrary Dimension Reed-Solomon Coding and Decoding for Extended - - PowerPoint PPT Presentation

▶

Nov 23, 2023 109 likes •243 views

Arbitrary Dimension Reed-Solomon Coding and Decoding for Extended RAID on GPUs Matthew Curry, H. Lee Ward, Anthony Skjellum, and Ron Brightwell University of Alabama at Birmingham Sandia National Laboratories November 17th, 2008 Matthew Curry,

SLIDE 1

Arbitrary Dimension Reed-Solomon Coding and Decoding for Extended RAID on GPUs

Matthew Curry, H. Lee Ward, Anthony Skjellum, and Ron Brightwell

University of Alabama at Birmingham Sandia National Laboratories

November 17th, 2008

Matthew Curry, et. al (UAB/SNL) GPU RAID November 17th, 2008 1 / 11

SLIDE 2

The Need for More Reliable RAID

Lack of Failure Prediction

◮ SMART ◮ MTTF

Larger Disks

◮ Stagnating Speeds ◮ Bit-Error Rates

Correlated Failures

◮ Batch-Correlated Failures ◮ Environment-Related Failures Matthew Curry, et. al (UAB/SNL) GPU RAID November 17th, 2008 2 / 11

SLIDE 3

Current Method: Nested RAID

Stripe data over several RAID arrays

◮ RAID 1 + 0: Stripe over multiple RAID 1 sets ◮ RAID 5 + 0: Stripe over multiple RAID 5 sets ◮ RAID 6 + 0: Stripe over multiple RAID 6 sets

Reliability is marginally improved over non-“+0” variants, while requiring significantly more hardware.

Matthew Curry, et. al (UAB/SNL) GPU RAID November 17th, 2008 3 / 11

SLIDE 4

Enabling RAID N+3 and Beyond

Need a fast method of creating arbitrary amounts of parity Reed-Solomon Coding is an obvious solution, but performance is lacking On an x86-based CPU, performance is limited to approximately 90 MB/s per core to do n + 3 parity Main limitation: A lack of the ability to do parallel table lookups, a crucial optimization for Reed-Solomon coding

Matthew Curry, et. al (UAB/SNL) GPU RAID November 17th, 2008 4 / 11

SLIDE 5

GPU Architecture

Figure: G80 Architecture

Matthew Curry, et. al (UAB/SNL) GPU RAID November 17th, 2008 5 / 11

SLIDE 6

Framing the Experiment

GPU Accumulate Buffer GPU Operate Buffer

...

Disk 0 Disk 1 Disk 2 Disk 3 Disk n+m

Driver copies write request data to accumulation buffers Buffers rotate as GPU finishes operating

n Operate Buffer

Retire write request and complete asynchronously

Operating System Kernel

Network Buffer Block Buffer Cache Network Buffer Driver (Finish Request)

Disk Writeout Buffer

iScsi Request iScsi Reply Network Packet Network Packet Contents of Operate Buffer Transferred to GPU Memory

GPU (Parity Calculation Performed) Matthew Curry, et. al (UAB/SNL) GPU RAID November 17th, 2008 6 / 11

SLIDE 7

Generation Performance

200 400 600 800 1000 1200 1400 1600 500 1000 1500 2000 Throughput (MB/s) Size of Input (KB) Parity Generation Performance NVIDIA 260: 13+3 NVIDIA 260: 29+3 Core 2 Duo: 13+3 Core 2 Duo: 29+3 Matthew Curry, et. al (UAB/SNL) GPU RAID November 17th, 2008 7 / 11

SLIDE 8

Recovery Performance

100 200 300 400 500 600 700 800 500 1000 1500 2000 Throughput (MB/s) Size of Input (KB) Data Recovery Performance NVIDIA 260: 13+3 NVIDIA 260: 29+3 Core 2 Duo: 13+3 Core 2 Duo: 29+3 Matthew Curry, et. al (UAB/SNL) GPU RAID November 17th, 2008 8 / 11

SLIDE 9

Percentage of Time in PCI Transfer

0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 500 1000 1500 2000 Percentage Size of Input (KB) Percentage of Time Performing PCI Transfer 13+3 29+3 Matthew Curry, et. al (UAB/SNL) GPU RAID November 17th, 2008 9 / 11

SLIDE 10

Conclusions

A $300 GPU can support the workload of a sizable RAID array that can support any three disks failing.

◮ 16-disk array at 100 MB/s per disk (vs. 7 for CPU) ◮ 32-disk array at 51 MB/s per disk (vs. 4 for CPU)

PCI-Express transfers can be fully hidden by the computation when done in parallel Future work includes building a working RAID system which includes this component (which will be available soon).

Matthew Curry, et. al (UAB/SNL) GPU RAID November 17th, 2008 10 / 11

SLIDE 11

Thank you.

Matthew Curry curryml@cis.uab.edu

Matthew Curry, et. al (UAB/SNL) GPU RAID November 17th, 2008 11 / 11