2011-12-22 12:00 PM
When running sysstat, my Disk Utilization numbers continue to be very high. My understanding of the DIsk Util field in sysstat is that it is reporting only the utilization of the busiest disk during a given interval. My question is, what are the best options for seeing a more complete disk utilization picture, perhaps disk utilization as a collection of all disks, or even more useful, which disk is the heaviest used.
I am hoping that with additional information I can try and bring disk utilization down, or at least understand what is causing the values to be so high.
2011-12-22 01:31 PM
You're correct in that sysstat just gives you the busiest disk (the same way it only shows you the busiest CPU). Anyway, invoke a priv set advanced, then before one of these events you're seeing occurs (if you can predict it and it's not just all the time) invoke the following:
Wait a period of time (usually more than a minute or so) relevant to the operation you're trying to monitor and then type the following:
Make sure your client buffer is set to a bunch of lines (I keep mine around the 2K mark) and then scroll up to the disk/aggregate section. This will show detailed information on each individual disk within each plex that makes up your aggregate. If you can tie an aggregate back to a particular operation and then see exactly which spindles are busy you'll know what client/process might be causing your busy disks.
If no particular process seems to be the culprit (you're not seeing high network I/O correlating with the busy disk for example) look to see if you have dedupes running which can be very disk intensive.
Hope that helps,
2011-12-23 12:26 AM
yea, he's right, statit is the one to see the ut% of each one. and u need to specify how long is the sample. cause all values in it are average. so longer sample, u may not see the peak ut%.