We’re doing some performance testing with a customer right now using solid state drives (SSD). The customer currently has 9x 146GB drives configured as R5 8+1. Marketing numbers will typically state between 15x to 30x IOPS performance over 15k fibre channel drives. We all know real world will typically yield less. Below is a table of tests run with iometer.
|Profile||IOPS||MBps||Response||Cpu||Drive Count||IOPS per Drive||MBps per Drive||IOPS Comparison to 15k|
|2KB and 4KB blocks 50/50 split – sequential||10973||32||185||6.4||9||1219.22||6.77|
|2KB and 4KB blocks 50/50 split – random||13996||31||145||8.6||9||1555.11||8.64|
|1M blocks sequential||264||264||5267||1||9||29.33|
|10MB blocks sequential||28.15||281||38911||1||9||31.22|
From an IOPS perspective, we got 6.7x to 8.6x improvement, assuming 180 IOPS for 15k drives. While it’s not 15x or 30x, it’s still impressive. You’ll note that the response time increases drastically for the larger block I/O and that CPU utilization climbs with the smaller block I/O, as you’d expect.
The numbers say it all.