Sign in to follow this  
shokunin

How to monitor average queue depth?

Recommended Posts

Hi all,

I'm trying to analyze how my usage patterns queue depth looks like, is Perfmon an accurate guide to monitor average queue depths on one of my physical disks? It doesn't seem to be correct as when I'm doing some heavy par'ing or other video work the queue depth ranges from 60-300+ (12 x WD4000YR on Areca Raid 6) which is way off the charts of other benchmarks looking at queue depths of 8, 16, 32 etc...

Thanks

Glenn

Share this post


Link to post
Share on other sites
Guest Eugene

Are you sure you're not getting confused by the y-axis scale? Try switching from "graph" to "report" mode...

Share this post


Link to post
Share on other sites

Looks like I was using the wrong metric, I was using avg. queue length instead of current disk queue length. If I use current disk queue length on my raid array it zig zags between 16-40+.

Share this post


Link to post
Share on other sites
Are you sure you're not getting confused by the y-axis scale? Try switching from "graph" to "report" mode...

Hi Eugene,

Interesting, so when I switch to report mode, I'm basically all under 1 queue? it ranges .3 to 3.8 is that correct? can you have a queue depth greater than zero but less and one? Please excuse my ignorance on disk monitoring but am trying to determine what is my usage pattern and what disk strategy would benefit me most.

Thanks.

Glenn

Edited by shokunin

Share this post


Link to post
Share on other sites
Guest Eugene
Interesting, so when I switch to report mode, I'm basically all under 1 queue? it ranges .3 to 3.8 is that correct? can you have a queue depth greater than zero but less and one? Please excuse my ignorance on disk monitoring but am trying to determine what is my usage pattern and what disk strategy would benefit me most.

Thanks.

Glenn

For average depths, yes... in fact, the vast majority of all single-user loads will average under 1. "Current Queue Depth" sticks to integers only. Perfmon's "scale" display on the default "graph" unfortunately tends to lead many users astray into thinking that they encounter massive queue depths. Instantaneous queue depths can hit 50+ VERY BRIEFLY when launching an application, but it falls down back to 1-2 at most within a second.

The best way to "audit" and verify that you're seeing what you think you do is to use a program that can generate and maintain a known concurrency of I/Os... IOMeter is an easily obtained one. Set queue depths to, say, 4, and watch how perfmon's Report Mode indeed reports a current and average queue depth of 4, not 100+ as the "graph" display would lead you to believe.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
Sign in to follow this