Performance Advisor backend metrics

[Chad Sellars]

Does anyone know where to find a detailed description of the metrics in performance advisor? The tables in the manual do not contain this information. I/O metrics are for the front end and are not as good as backend metrics for identifying hot raidgroups. Backend metrics are expressed in tracks per second. Why? I am familiar with IOPs, not tracks per second. Or, is this just the actuator seek rate? The actuator seek rate is really what limits random R/W. However, for sequential I/O, the sectors per track dominate the throughput.

Does anyone know what is the maximum tracks/sec is for a 10K drive? Perhaps, that depends on the ratio of random to sequential. For purely sequential, guess it would be very close to the reciprocal of the head switch time. Track and cylinder skew reduce these values far below a random seek. I think a head switch is around 1 mS. That is a 1000 tracks per second. For random I/0, a 10 K drive can do about 120 seeks/sec.

So what value are these track per second numbers in PA if they are so different between random and sequential? Who can make sense of them? How can I tell when a raidgroup is too hot?