Disk IO activity, LVM and glance.

[Tim Rotunda]

I am running a VA-7100, with a single controller, which has 3 LUN???s on 9 36GB disks, all configured into a single lvol. The second two LUN???s were added after the first one was nearly full. Glance shows that the first device (LUN0) has 99% of the activity, the second device (LUN1) with only about 1% and the third device (LUN2) virtually idle. I understand why this is but I am trying to understand what effect, if any, it is having on system performance. Below illustrates my point. Index 3, 4 and 5 are the VA-7100. Glance shows utilization on device 3 as 100%, but that doesn???t change unless activity drops below 2KB/Sec or so. I am thinking so what, Glance says my Disk Util is at 100%, but the other data doesn???t seem to indicate a bottleneck on Disk IO. Any thoughts?

Thanks very much,
Tim


Idx Device Util Qlen KB/Sec Logl IO Phys IO
--------------------------------------------------------------------------------
1 0/0/2/0.0.0 91/ 80 0.2 655.1/ 723.0 na/ na 150.7/137.9
2 0/0/2/0.2.0 72/ 60 0.2 564.8/ 605.9 na/ na 126.5/108.5
3 0/12/0/0.8.0.110.0 100/100 5.0 4329.7/ 4874.4 na/ na 305.6/310.7
4 0/12/0/0.8.0.110.0 15/ 11 0.3 136.5/ 237.8 na/ na 8.7/ 9.8
5 0/12/0/0.8.0.110.0 0/ 0 0.0 0.0/ 0.0 na/ na 0.0/ 0.0
6 0/0/1/1.0.0 0/ 1 0.0 0.0/ 6.7 na/ na 0.0/ 0.9
7 0/0/1/1.2.0 0/ 0 0.0 0.0/ 0.0 na/ na 0.0/ 0.0

[Vincent Fleming · Accepted solution]

Utilization or % busy statistics are not accurate with cached disk arrays, so don't worry about it.

If you want to make sure you have no bottleneck on the array, look at the service times. Another good indicator is CPU WIO.

It makes no difference how many LUNs you have on a VA7100 - your LUNs are all striped across all 9 of your drives. So, don't worry about the VA's configuration.

Good luck!

[Martin Johnson]

As mentioned previously, % utilization is not a good measure of a bottleneck. Check the I/O queue lengths instead.

HTH
Marty

[Sridhar Bhaskarla]

Hi,

As indicated above, %util is not valid for raid systems. Since you have configured only one file system on this LUN, it is used all the time hence the %util show it as 100%.

Sometimes, sar may help better than glance.

Do a sar 2 20 and observe %wio. If it is more than 20, then there is a chance that the system is waiting on the disk.

Do a sar -d 2 20 and observe avqueue and avserv. A high avserv indicates that the service times are high and there is a bottleneck particularly if the corresponding avqueue is looking suspecious.

Also in glance type "B" and look at "DISK IO" metric. This is for Glbal wait states and you should see only a little % waiting on disk IO.



-Sri