Server Fault is a question and answer site for system and network administrators. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I have a server with room for a total of 6 disks. Ultimately, it will be running VMs hosting various services, such as Exchange, and others. My idea is to get the stores and transaction logs on seperate spindles, and possibly even the OS itself on its own spindle. The server is mission-critical, but does not need to have 24/7 uptime. It will be running vmware, though all of our other servers are running Hyper-V as a service. How should I divide up the disks?

Do I do three RAID-1 arrays, for maximum spindle sets?

Do I do a RAID-1 and a RAID-10? RAID-10 has higher performance than just a plain stripe/mirror AFAIK, but if my workload performance is dependent on separate spindles, could this be unoptimal?

Should I opt for two RAID-5 sets? As far as I know, the only benefit to this would be more storage space. Are there other benefits I'm missing?

Or should I just RAID-6 the whole goddamn thing? ;)

Edit: The comment on the OP of this post seems to follow my line of thought--that if IOPS are important, more spindles are preferred. But I'd like to hear some other opinions too.

share|improve this question
Reading this, my initial reaction is that this is a bad idea. You're right that mailbox databases and transaction logs should be on their own disks, as should the OS really. You've pretty much used all your disks on this one VM if you follow recommended practices. You should be aware that performance on all VMs on this box may be adversely affected. Obviously your mileage may vary, but it certainly sounds like an I/O nightmare. – Ben Pilbrow Aug 22 '11 at 18:44
It's better than our current situation, where we have 3-4 VMs running off a single RAID array. The Exchange server only has about 200 users, so it's not a high-capacity system. Other VMs could still go on the other disks beside Exchange. – Bigbio2002 Aug 22 '11 at 18:52
up vote 4 down vote accepted

If this were a bare-metal Exchange or DB box then I'd absolutely recommend creating separate RAID arrays but given this is for VMs that changes things. Personally I'd just create big R10 array, stick the hypervisor and the datastores/VM-volumes on the same array (different partitions) and leave it at that. You'll get pretty good performance out of that and it'll be far easier to deal with and you run less of a risk of one array running out of space when there's loads left on the others.

You really could do with letting us know what hypervisor you intend to use though.

share|improve this answer
It'll be vmware. I added it to the OP. – Bigbio2002 Aug 22 '11 at 18:54
VMWare is a company not a product, which of their products are you talking about? – Chopper3 Aug 22 '11 at 18:56
VMware ESX Server 4.1. – Bigbio2002 Aug 22 '11 at 19:02
Ah yes, same as I use - yeah, create one big R10 array but then create multiple logical drives of 2GB exactly, VMFS3 doesn't deal with >2TB partitions ok. – Chopper3 Aug 22 '11 at 19:05

Why to get RAID 10 over RAID 6 :

  • Better write performance. RAID 1+0 imposes only a 2x write performance hit.
  • Faster rebuild speed. Rebuilding a failed disk that takes part in a mirror is a much faster process than rebuilding a failed disk from a RAID 6 array.
  • Performance degradation during rebuild process is minimal. When a RAID 6 disk fails, the rebuild process can have a seriously negative impact on overall storage performance due to the need to recalculate parity. With RAID 10, re-establishing a broken mirror is a relatively behind-the-scenes process.

from techrepublic

share|improve this answer
I wasn't seriously considering RAID-6, I figured I'd just throw it into the list of options to be fair. – Bigbio2002 Aug 22 '11 at 19:02
Bare in mind of course that ESX/ESXi has zero software RAID capabilities, therefore you'll HAVE to use a hardware RAID controller, at which point the 'double-writing' that Lucas mentions will be entirely handled by the controller so won't add extra load to the CPU or bus - another reason why R10 is what you need :) – Chopper3 Aug 22 '11 at 19:18

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.