HPE Storage Users Group

With the 7000 series AO and DO are in the same license suite so if you have one you have both. As far as the log volume versus database, if anything I'd put the the database (and especially tempdb) on RAID10 and the logs on RAID5, the reason being that the database will have some small writes (8kb) while the logs will always write 64kb or greater pieces so you can set your stripe size to 64kb and be assured of full stripe writes.

Which is as I thought but I've been reading some more and although this best practice doc is a while old, it is still relevant and it says this:

Microsoft Exchange Server 2007 Best Practices Using 3PAR Utility Storage

I know the 3PAR throws most things out the window but that seems counter intuitive since RAID10=best for writes, RAID5=slight advantage in reads so wouldn't it make sense to put the DBs on the RAID5 and the Logs on RAID10? Also, why can't I set the stripe size to 64kb in RAID10? The option is there to do so.

As long as you are doing full stripe writes there is basically no performance difference between RAID5 and RAID10, there's a very slight bit of latency but it's orders of magnitude lower than the disk latency so it's essentially noise. Now when it comes to the database storage, there you will get 8KB writes, a write that is smaller than the stripe size incurs a significant penalty in RAID5 (basically you have to read the current stripe, modify it, then write the new stripe out).

ok. so pulling this together, I'm limited to RAID1 across these 2 cages as you pointed out in my other post. So now I'm going with one RAID1 CPG with the default 256kb step size for general VVs and two RAID1 CPGs ,one with odd disks, one with even disks and both with 64kb step sizes. One will hold SQL and Exchange Database VVs and the other will hold the Log VVs. I'm hoping this will combine the database optimisation with the 3PAR best practice recommended in that Exchange guide for keeping the logs and databases on separate disks. I'm still not sure of the benefit of this, especially with so few spindles in my array so I'll have to see what IO testing I can do to see which offers greater benefit. More shared spindles or fewer isolated spindles.