up vote 0 down vote favorite
share [g+] share [fb]

I'm sure this is going to be very dependent on the harddrives and other hardware that we have in our machine, but I have no idea how to read these graphs.

This graph looks nice and low.

enter image description here

This one looks scary as hell.

enter image description here

I know that the scale is different, but how do I know when things are starting to go wrong as far as the disk actually struggling to keep up?

link|improve this question
89% accept rate
The attachments aren't showing up. – Nixphoe Jun 15 '11 at 21:29
I'm not sure why not. I viewed the source and they are actually coming from the web (not my local machine). To explain, basically, when Resource Monitor uses 100KB/sec scale for disk utilization, it looks scary (stuff is pegged). When the scale is 1MB/sec, the graph looks nice and low. How can I tell when things are wrong? – Matt Dawdy Jun 15 '11 at 21:31
feedback

1 Answer

up vote 2 down vote accepted

If you only need to know if your disk is bottlenecking something, the basic parameters for disk performance are queue length (you might want to look at read and write queues separately, sometimes this is insightful), the disk idle time percentage and possibly the request service time. The queue length value tells you how many requests for disk data are outstanding but not yet serviced because the disk is busy doing other things. If you see anything there that is larger than the number of spindles (i.e. the number of disks in your disk array if you have one or simply "1" if you only have a single disk to monitor) for a prolonged period of time, your applications are certainly waiting for I/O a lot and you should look into it. You will see the disk idle time go down to 0% (or the disk busy time go up to 100%) simultaneously, because if there is always at least one request in the queue, then there is never time for the disk to be idle.

The resource monitor shows the disk's busy time percentage (blue lines in the graph) as well as the queue length and the throughput. Throughput is hardly of any value except that it is nice to look at. None of the given graphs looks "scary" in the "my system is under high load" way - the disks are mostly idle (blue line crawling at the bottom, mean disk queue length << 0.05)

link|improve this answer
Thank you. I honestly had never noticed the blue line at the bottom of the Queue Length graph. I'm just trained that when things get to the top, it's bad. In this case, from what you have said, all is doing just fine. – Matt Dawdy Jun 15 '11 at 21:46
The resource monitor uses variable scaling. No matter what you do, the graphs will always go "to the top" as the scale will be readjusted accordingly. You need to pay special attention to the scale values to get any meaning out of the graphs. – syneticon-dj Jan 4 at 23:55
That was my point. If the graph scale goes from 1 to 10, is that bad? I mean, does it scale up to 100 and 1000, and THEN I should be worried? It's no longer an issue anymore. – Matt Dawdy Jan 5 at 0:14
1  
There usually are rule-of-thumb guidelines for different counters and different workloads. But of course nothing beats your own experience (like in "the last time I tried to work on a machine with the disk queue length of 10 it really really sucked") since you know your machine's limits best (a queue length of 10 would be unbearable to work with on a single-disk workstation, but pose a rather low load in an 80-disk SAN appliance) – syneticon-dj Jan 5 at 0:22
feedback

Your Answer

 
or
required, but never shown
discard

Not the answer you're looking for? Browse other questions tagged or ask your own question.