I've got an IBM x3550 M2s (2 x 4w nehalem 2.93Ghz, 32GB ddr3 1333GHz, with 1 x 300GB 10K disk SATA). I'm running an application which is a heavy database (DB2) workload with a lot of writes. I've collected iostat data on the run with 10 second intervals. The average (r/s + w/s) iops is 744 iops and there is a max of 1733 iops. For the 1733 point, 1731 of the iops are writes, the wsec/s is 15934, avgqu_sz is 3.9, await is 2.25, svctm is .56. Everything I've read says a single drive can only handle between 120-150 iops. So, how could the iops average be over 700 and be able to reach over 1700? Is this the affects of caching, could the mix of sequential and random IO explain this, is the iops only an indication of the requests going to the drive--while the 120-150 range is what a drive is really handling so everything else is being queued?

I don't think you're going to get an accurate picture that way.

There's caching going on all over the place. The kernel caches, and writes to the disk cache of the drive. Probably the controller caches as well. It has to select the drive each time, as sata allows more than 1 drive per connection. Perhaps talk about partitions and directories. Go figure.

What does hdparm -tT say?