RAID 0 128K Sequential Read/Write
One interesting caveat of RAID testing is the performance impact of different RAID settings. This is very clear with read caching enabled and disabled on the ASR-72405 RAID controller. We observe that the top precondition chart, with read caching disabled, performs with a typical I/O pattern, though the scaling is not very impressive with direct writes to the SSDs.
Enabling read caching provides a huge boost in performance, as shown in the second graph. It also brings out an interesting lattice-like pattern of performance. In our testing for the article we are focusing on the performance of the Intel SSDs, so we conduct all testing, except this small example, without read caching. We will take a deeper look at the varying performance with different caching settings in future RAID controller articles.
The 128K read sequential speeds reflects the maximum sequential throughput of the SSD using a realistic file size encountered in an enterprise scenario. The large array tops out just shy of 3,000 MB/s, while the smaller array peaks at 1,675 MB/s.
128k read latency results indicates the smaller array is more nimble at much lower Queue Depths, where its latency is low and any increasing load does not yield a tangible increase in bandwidth. The larger array reaches the best IOPS/latency ratio at QD64.
The 8-drive array peaks at 1,678 MB/s at QD256, while the 4-drive array provides 857 MB/s at the same Queue Depth. While this is near-linear scaling, there is quite a bit of speed left on the table with each drive only providing 209 MB/s, while they are rated for up to 365 MB/s each.