* w/s: This is the number of write requests submitted to the device per second. My testing shows good write-per-second correlation with Oracle, when adding together the database writer IO operations from the v$filestat column phywrts and log writer requests from the v$sysstat statistic redo_writes. It is troubling, however, that adding all wait event write-related occurrences fell well short of the actual iostat r/s and v$filestat and v$systats numbers.

* await: This is the average IO request response time in milliseconds. While clearly mislabeled, the manual page and observations show the await column includes both the true service time and queue time. Before I begin an IO analysis or make this statement to a customer, I always double-check the manual page for the particular Linux or Unix implementation. Again, my rule of thumb for acceptable response time is 10 ms.

* %util: This is the average device utilization. Modern IO subsystems can sustain utilizations up to and possibly over 65%. While I use a high utilization to spot potential issues, I also make sure the response time is an issue before claiming there is a problem. I have seen devices 90% busy and still servicing IO requests in under 5 ms. The arrival pattern and IO request service time play a significant role in the resulting device utilization.