Tuesday, 24 February 2015

Measuring disk performance using performance monitor on Server 2012

The following counters each offer you vital information when attempting to diagnose disk perfomance issues, with either a LUN, RAID Array or simply a single disk (taken from TechNet):

%Disk Read Time, %Disk Write Time, %Disk Time, %Idle Time: All of these can be important, but keep in mind that they are only reliable when dealing with single disks. Having disks in RAID or a SAN setup can make these numbers inaccurate.

Average Disk Queue Length: A good counter to monitor if requests are backed up on your disk. Any sustained number higher than 2 is considered problematic. However, just like the counters above, this one is not considered reliable except when dealing with single physical disk per volume configurations.

Average Disk Second/Read, Average Disk Second/Write, Average Disk Second/Transfer: These are probably my favorite disk counters. They tell you how long, in seconds, a given read or write request is taking. (Transfer is considered a Read\Write round trip, so is pretty much what you get if you add the other two.) These counters tell you real numbers that deal directly with disk performance. For instance, if you see that your Average Disk Seconds/Write is hitting .200 sustained, that tells you that your write requests are taking a full 200 ms to complete. Since modern disks are typically rated at well under 10 ms random access time, any number much higher than that is problematic. Short spikes are okay, but long rises on any of these numbers tell you that the disk or disk subsystem simply is not keeping up with load. The good thing is, these being real numbers, the number of disks or how they are configured is really not important. High numbers equate to bad performance regardless.

Current Disk Queue Length: Don’t bother with this counter. It measures instantaneous disk queue numbers, and as such is subject to extreme variance. You can get much more useful numbers by using the Average Disk Queue Length counters mentioned above.

Disk Bytes/Second, Disk Read Bytes/Second, Disk Write Bytes/Second: These counters tell you how much data is being read from or written to your disks being monitored. The number by itself doesn’t mean anything, but in combination with other counters can be telling. For instance, let’s say it is reporting that you are writing 100 MB/second to the disk. Is this too much? Not if overall performance is still good. If however, your disk access times are getting high, you can see if you can correlate this to the load reported by this counter and calculate exactly how much your disks can handle before becoming slow.

Disk Reads/Second, Disk Writes/Second, Disk Transfers/Second: These counters tell you how many operations are occurring per second, but not how much as being transferred. It is possible for instance to have a large amount of very small requests causing performance issues, but overall throughput may not be high enough to seem problematic. This counter can be used in the same way as the Disk Bytes/Second counters, but measure operations instead of throughput.


Post a Comment