12H AutoRaid, checking load on each controller

[Bernie Vande Griend]

I have a 12H AutoRaid that I inherited. It was set up with 8 LUNs with 5 volume groups, both controllers were set up. It was not striped between controllers, though, basically each volume group just has a primary and alternate path. I know the best thing would be to redo the setup of the AutoRaid and stripe it between controllers, but the downtime in doing that is not acceptable.

Currently one volume group is set to use the path of one controller and the other 4 VGs are using the other controller path. I want to spread the load out better. I know how to do this, but is there a way to measure the load on each controller and track the performance gains after switching one of the volume groups? I could add up the IO of each LUN, but I'm looking for a better, more accurate way to do this either through Measureware/Glance or the AutoRaid software.
Thanks and I will assign points. -Bernie

[Bob Inglis]

I have worked with people who have used GLANCE to identify a disk bottle-neck.

[Bernie Vande Griend]

Thanks for the reply. We are already using glance and other tools to look at the IO of the disks. We want to measure the IO of the controller and look at its limitations.

[Curt Thompson]

Hello Bernie,

Would 'arraydsp -m' (for performance metrics) be of some help?

arraydsp -m starttime endtime array-id

Or, 'arraydsp -r' for Autoraid-generated recommendations.

Good Luck,
Curt

[Bill McNAMARA_1]

I don't think you can because the controller itself is the targettable device.
It's the same problem trying to find the load on any scsi targettable device.
iostat for example is only going to class on disk io.. the d field of the sdisk dev file.

I would imagine the only way to get any usefull information related to the controller is via arraydsp -c
The performance metrics and recommendations I have found to be rather useless. You should concentrate more on the disk/lun io through the controllers rather than the controller itself, afterall there's not mich you can do.. everything is mirrored even on the controllers.
(apart from reads)

You can play around with the arramgr -J option but comprimise true HA versus performance.
LVM balancing is the true analysis method, using perfview and glance and so on.. you then decide to access on the other bus via the other controller.

Later,
Bill

[Bernie Vande Griend]

Thanks guys. Curt, I've already looked at that, but it doesn't tell me the load through the controller either. -r option basically tells me everything is ok.
Bill, you basically are confirming what I expected, that there really is no way to track and measure the load of the controller devices, just the disk devices themselves. And I agree that the metrics don't do me a whole lot of good either. I will just concentrate of the disks themselves, checking the load on each and making sure it is split as evenly as possible between the 2 controller until I have an opportunity to do this the right way and make both controllers evenly used all the time.