Disk Arrays
cancel
Showing results for 
Search instead for 
Did you mean: 

Auto Raid 12H Performance Questions

Robert Landstrum
Occasional Advisor

Auto Raid 12H Performance Questions

I have an Auto Raid 12H, and currently I only have one LUN in production. When Glance Plus is running, it will show an alarm that there is a 100% probability that there is a disk bottleneck problem. And the disk graph will show 100% utilitzation on the primary controller for the LUN. But none of my users are reporting any slowdowns.

I was wandering if this was really a problem with Glance Plus or if I could be having disk bottleneck problems?

On one hand I am not concerned because the users are not reporting any problems, but on the other hand I am concerned that this could be a problem down the road.

Thanks in advance for your help.
If I thought the answer was easy, I would have called HP.
4 REPLIES
Andreas Voss
Honored Contributor

Re: Auto Raid 12H Performance Questions

Patrick Wallek
Honored Contributor

Re: Auto Raid 12H Performance Questions

I have two AutoRaid 12Hs and I personally don't really like them, but anyway, here is what I have done to try and remedy the performance issues.

I have two controllers in the AutoRaid attached to 2 different SCSI buses so I have dual paths to the disk arrays. I then split up the array into 4 LUNs. I have the primary path on 2 of the LUNs set to SCSI Bus 1 and the primary path on the other 2 LUNs set to Scsi Bus 2. I then created the VG on the array using those 4 LUNs so that the I/O was split between the 2 different SCSI Buses and the 2 controllers on the array. It helped. The only problem with doing that is that it required that the data be backed up and the 12H totally reconfigured and the data restored.
Paul McCleary
Honored Contributor

Re: Auto Raid 12H Performance Questions

A better way of doing what Patrick is talking about is to stripe down the 2 controllers. For every volume group u create you use 2 LUNs. E.g. If you want 10GB, u would create lun 0 and lun 1 each at 5GB, an example ioscan would show you c5t2d0, c5t2d1, c6t3d0, c6t3d1. Create your vg using c5t2d0 & c6t3d1, vgextend it to include the alternate paths. When you create your logical volumes use:

lvcreate -i2 -I64 -L 10000 -n lveg /dev/vgeg

This tells it to stripe your logical volume across the 2 luns so it sends 64k chunks down each controller to your logical volume.

You can check it by doing an lvdisplay of the lv and it'll show you it striped across the 2 luns.

Of course you will need to backup your data first.

Patrick probably dilikes the AutoRAID because he's a DB and prefers to know what spindles his logs and dbspaces are going onto. The AutoRAID takes this mgnt overhead from u and if setup correctly it can be very performant, it u don't like that u should buy jbod.

Hope this helps, Paul
Patrick Wallek
Honored Contributor

Re: Auto Raid 12H Performance Questions

Paul,

I am a SysAdmin, not a DBA. The reason I dislike the 12H arrays is because of the architecture. The fact that the disk drives in the array are SE scsi and controllers are FWD scsi is crazy. Granted, with haveing I/O going to multiple disks from the controllers this shouldn't be a problem, but to me it's just asking for trouble.

Unfortunately, to me anyway, I inherited the 12H's when I took over the Sys Admin job on these systems. All the disk I have bought for these systems since then has been JBOD.