Mirror Disk performance

[Michael Barron]

Ok, let me first explain my setup.
We are running a partition on our Superdome connected to an XP-12000 disk array. We have all of our LUNS nmirrored to a different array, same type XP-12000. The server is being used for a proprietary DB. My question is, I know reads will be fine, but what about writes? Is there much degredation? If so, can anyone point me to a doc that shows some numbers, i.e, 10%, 20% or 30% degredation? Thanks Much

[Michael Steele_2]

hi Michael:

Since I support Oracle most of the time I usually refer to their recommendations. For example, this url:

http://www.oracle-base.com/articles/misc/RAID.php

[Trevor Roddam_1]

Hello.

I'm assuming you're asking about any performance issues with using LVM RAID 1 between the arrays and there also being RAID within the arrays?

The answer is it depends.....

Some more assumptions:
That the arrays are identical as are the LUNs.
Each array is via the same FC card and Switch
Array to Array mirroring is used instead of Array based replication to reduce cost, increase flexibility and possibly uptime.

In theory there shouldn't be any difference in performance between the arrays. But this rarely happens as there are other hosts using them in different manners.

There are no freely available documents describing this as vendors don't promote it as the preferred solution. It is normally used to have an HP array and an EMC (for example) supporting the same data incase there is a flaw in Array.

The only performance issue will be the differential between the arrays. Going back to mirroring lvol's between disks that performed differently EG 2GB 7200 rpm HDD to an 18GB 15K rpm HDD. Obviously the best write performance is that of the slowest device. Same here.

Is there an performance hit mirroring between the arrays instead of using array based replication?
Not if they are set up correctly. But setting it up correctly is complex.

Hope this helps.
Trevor.