I'm performing some benchmarking with iometer and our SAN (NexentaStor) and I'm having some puzzling results..
I have a production SAN that's a raid2z pool with (8)15k SAS drives and mirrored 32GB Intel X25-E SSDs for ZIL.
Test Name | Avg Iops | Avg MBps | Avg Resp (ms) | CPU | Queue Depth |
---|---|---|---|---|---|
100% Read | 3426 | 107 | 17 | ? | ? |
RealLife | 2944 | 23 | 14 | 30% | 52 |
Max Throughput | 3848 | 120 | 17 | 14% | 72 |
Random | 2709 | 21 | 15 | 40% | 56 |
Next, I have totally different lab SAN that I'm going to be turning into production, that's a RAID 10 with (8)15k SAS and a 32GB Intel X25-E SSD for ZIL
Test Name | Avg Iops | Avg MBps | Avg Resp (ms) | CPU | Queue Depth |
---|---|---|---|---|---|
100% Read | 3495 | 109 | 17 | ? | ? |
RealLife | 10294 | 80 | 5 | 41% | 64 |
Max Throughput | 4164 | 130 | 5 | 15% | 64 |
Random | 11141.2 | 87 | 4 | 37% | 56 |
Also on this lab box I have a CHEAP RAID 10 with (4) 7.2k SATA drives with NO ZIL that I'm going to be using for ISOs and low priority VMs (Tier3 basically).
Test Name | Avg Iops | Avg MBps | Avg Resp (ms) | CPU | Queue Depth |
---|---|---|---|---|---|
100% Read | 3498 | 109 | 4.5 | 11% | 16 |
RealLife | 2380 | 18 | 21 | 31% | 60 |
Max Throughput | 1951 | 60 | 28 | 26% | 64 |
Random | 2578 | 20 | 18 | 33% | 56 |
So, what's sticking out like a sore thumb for me is that the 100% Read tests for ALL of these tests are pretty much exactly the same! I could possibly see how the same 8x15k volumes (raid2z vs RAID10) *might* be the same, but vs the 7.2k SATA drives??
More details happily given if needed.