up vote 1 down vote favorite
share [g+] share [fb]

If I only had 1 pair of 15,000 RPM SAS drives in RAID 0 to dedicate to a VM box, would it be best to

a) put the HOST OS on the raid drives and put the VM's on a single SATA 10,000 RPM drive

b) put the VM's on the raid drives and the OS on the SATA 10k drive.

c) Put both the OS and the VM's on the SAS RAID and forget about the SATA 10k.

If so, why?

Not asking for benchmarks here, just some logic of what would theoretically be best on a small-scale low-traffic VM box.

link|improve this question
63% accept rate
I would sell all of the 15K and 10K drives and buy one or more Intel (or better) SSDs. – Miles Erickson Jul 29 '10 at 21:53
feedback

3 Answers

up vote 4 down vote

All on the 15k SAS drives, but in RAID-1 not RAID-0. The host shouldn't be doing anything much, so won't use it's disk IO much, while the guests doing the work will need more IO, and a single drive isn't much use for anything.

link|improve this answer
Thanks! Why RAID-1 and not RAID-0 though? If I don't really need 100% uptime and I have automated image backups, can't I afford RAID-0? – Matias Nino May 2 '09 at 6:38
7  
If your data is really that unimportant to you that you're willing to wipe out the whole thing every time you lose a disk, go nuts. I would consider the use of RAID-0 for anything more than a large scale cache to be bordering on negligence, though. Big disks are cheap, if you need the space, and the testing I've done doesn't show all that much of a performance benefit. – womble May 2 '09 at 9:01
+1 for RAID-1. It's a total waste of time rebuilding the system everytime a disk dies. – Edmund Tay May 2 '09 at 10:03
2  
Yes, because ZFS magically manages to provide infinite redundancy without any storage or performance overhead... – womble May 3 '09 at 5:05
1  
A year and a half later and this host is churning along with several VM's. The testing performance is top notch and all the time saved from the superior performance of RAID-0 will be well worth the 2 hours of downtime if and when a drive fails. – Matias Nino Jan 2 '11 at 6:08
show 3 more comments
feedback
up vote 1 down vote

I would:

  • Put the OS on the RAID-0 15k drives
  • Distribute some virtual disks on the 15k drives
  • Distribute other virtual disks on the 10k drive

I/O is truly the enemy of virtualization, so using as many spindles as possible is the win here.

If you can segregate your usage in any meaningful way (i.e. VMs "A" and "B" usually access the disks concurrently) then use that knowledge to intelligently choose which virtual disks to place on each drive.

link|improve this answer
Interesting answer. Clearly distributing among the scare I/O is the major thing to go for here. Makes me wonder whether a VM on a separate 7200RPM single drive would beat out one on a 15K RAID volume shared with the OS. – Matias Nino May 4 '09 at 8:02
feedback
up vote 0 down vote

if space allows it put everything on mirrored sas drives - giving you security and io.

link|improve this answer
feedback

Your Answer

 
or
required, but never shown
discard

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