No Avg. Disk Queue Length

Sep 10, 2010 at 6:10 PM

Clint, I'm curious why the Avg. Disk Queue Length counter is not a default counter in the System Overview Template? Many tuning/performance experts always tout monitoring the Disk Queue Length. I’ve been adding the counter manually and allowing PAL to process all counters. I'd just like to hear your reason.


Sep 10, 2010 at 11:08 PM

Avg. Disk Queue Length is valuable only if you know how many physical spindles are behind the LUN. With disk virtualization as the dominate force in enterprise environments, it is sometimes impossible to know what the physical spindle count really is. We typically go with a threshold of Avg. Disk Queue Length of greater than 2 plus the number of spindles. Since PAL doesn't know how many physical spindles are behind the LUN, it can't establish a proper threshold for it. I am happy to add it as a Stats Only analysis, so we can at least chart it.

In the Vital Signs workshop that I teach, we go with disk response times. The reasoning is that no matter how much you virtualize the storage, if the response times are good, then great, but if response times are bad, then thats when it warrants more investigation.

Check out my blog entry where I talk about how to analyze enterprise disks hosted on a SAN:

With all of that said, I agree that we should at least have the stats of Avg. Disk Queue Length, so I will add it to the next release. Thanks for the feedback!

Sep 10, 2010 at 11:09 PM

I forgot to add the counters for disk response times:

\LogicalDisk(*)\Avg. Disk sec/Read

\LogicalDisk(*)\Avg. Disk sec/Write