looking for recommendations on ESX4 OS - VMFS version3:

RAID-5 : matching the stripe size with VMFS block size? (64K, 128K etc)

RAID controller options: "adaptive read ahead, write-back" on PERC 6i

i have read that smaller stipes are good for writes and larger for reads. Since this is virtual env, not sure whats good.

what impact if any has aligning VMFS & data disks in Virtual machines? Appreciate your comments.


I think that it's more important to know what the vm's will be doing. Are they heavy IO, especially writes?

ESX can be very heavy on disk IO, in fact it's usually the biggest bottleneck. You end up doing a lot of big file copies (creating new guests, cloning and vmotion) and you are combining the disk IO of several servers onto a envirmonment usually reserved for a single server.

SQL Server is particularly write heavy, so on that alone I'd try to use RAID 10.

  • mostly balanced reads, writes this is a test VM server. So not really many big file copies, no vmotion. it is the usual load of Windows OS plus some writes to SQL Server. – John-ZFS Apr 25 '10 at 14:50

If you're using 2003 server, make sure your NTFS partitions are aligned with the underlying storage. Look at http://www.nptech.com/blog/2009/03/04/how-to-align-virtual-machine-os-and-data-partitions-in-esx-server/ for one method. I tend to be lazy and partition 2003 server disks with a Vista, 7 or 2008 server install DVD ISO, and then populate the resulting partitions having booted from the 2003 install media.

If you're going to be running write heavy workloads, I'd benchmark RAID 5 against RAID 10 before you enter production. RAID 5 with multiple concurrent writes can be a performance challenge.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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