On Mon, Jul 11, 2011 at 02:15:04PM -0700, Matthew Dillon wrote: > :If I interpret the results correctly, I get 1000-1300 iops: > :With 16KB requests, the numbers are in the 1400-1600 tps range: > > This sounds reasonable for a RAID system. Each disk should be able to > do ~400 iops, but RAID won't scale 1:1 on the disk count. It would be > worth testing without the CAM fix to see if it in fact fixed anything > (then I can commit it).
I've rebooted the machine with a non-patched kernel. With 512B request, I get 1000-1400 iops: randread 1.000s 1068 loops = 936.391uS/loop randread 1.000s 1210 loops = 826.489uS/loop randread 1.000s 1116 loops = 896.104uS/loop randread 1.000s 1161 loops = 861.374uS/loop randread 1.000s 1223 loops = 817.704uS/loop randread 1.000s 1294 loops = 772.838uS/loop randread 1.000s 1407 loops = 710.763uS/loop randread 1.000s 1259 loops = 794.315uS/loop With 16KB requests, I get 1200-1600 iops: randread 1.000s 1443 loops = 693.033uS/loop randread 1.000s 1475 loops = 677.997uS/loop randread 1.000s 1559 loops = 641.467uS/loop randread 1.000s 1438 loops = 695.445uS/loop randread 1.000s 1260 loops = 793.685uS/loop randread 1.000s 1449 loops = 690.162uS/loop randread 1.000s 1465 loops = 682.627uS/loop randread 1.000s 1584 loops = 631.345uS/loop I'll say the arcmsr driver modification didn't change anything visible. IOPS performance is about the same in all cases. -- Francois Tigeot
