Atto/iStoragePro Test Results for RAID-0 Striping
iStoragePro with Atto R380 using Oct 26 2010 firmware
Scalability is near linear through 5 drives for a 128K interleave. A very small loss in scalability is seen at the 5-6 drive mark (when using a 64K interleave, the drop is even more pronounced at 5 drives).
Higher performance for RAID-0 striping (10% at most) can be achieved with the Jan 2010 firmware, but it didn’t feel appropriate to use 9-month-old firmware. According to Atto tech support, “the only known issue with performance is resolved by applying the firmware from January 2010”. No explanation given as to why current firmware would abandon a performance fix. I opted for the latest firmware, figuring that testing with shipping firmware was more appropriate than testing with old firmware. But both firmware versions severely underperform the potential of the Western Digital RE4 drives that I tested when 8 drives are used.
When I asked Atto tech support why I wasn’t seeing the ~ 1080MB/sec that could be expected, I did not get a satisfactory explanation. It’s plain to see that the the Atto R380 card is throttling the performance, that the drive speed is not the issue. Even with the January 2010 firmware, 8 drives topped out at 725/852 MB/sec, still flatlined, and well below potential.
If scalability were maintained out to 8 drives, performance should be 1080 MB/sec, well off this chart. Using 8 drives, 30% of the read speed is lost, and 40% of the write speed is lost. That’s a huge hit if one is looking for maximum RAID-0 performance, the only defensible reason besides capacity to choose RAID-0 over RAID-6.
With software RAID, I have observed near-linear performance using twelve (12) drives using eSATA. So any claim that it’s not realistic to get full performance is disingenuous coming from a company whose expertise falls in this area. Unless the full meaning is “using our card”. Perhaps the Atto R680 card, with twice the bandwidth, is capable of more linear performance.
I observed some performance unpredictably with all forms of RAID; at times performance would rise or fall for no apparent reasons, and stay that way consistently.
This graph shows that RAID-0 read performance can hit nearly 1000MB/sec on the read phase of a DiskTester fill-volume test, though the write performance is throttled at ~640MB/sec.
Immediately after obtaining these results, I ran a run-sequential test, and observed 20-30% lower speeds on a freshly-erased volume (but with no changes to the RAID set). Including lower read speeds than write speeds!!! There is something unpredictable with the Atto RAID card performance (RAID set was configured with SpeedRead=Always, Prefetch=0).