Tuesday, November 16, 2010

Testing to see if aligned partitions on ESX - an Introduction

I have been looking at documents that detail speed improvements with ESX where the partitions are aligned.  These documents detail and only deal with SAN paritions:

http://media.netapp.com/documents/tr_3593.pdf
http://www.vmware.com/pdf/vi_performance_tuning.pdf
 http://www.vmware.com/pdf/esx3_partition_align.pdf

Now I don't have access to a SAN but wanted to develop and test this with my Direct Attached Storage (DAS) for my ESXi host.  Here is the configuration of my ESX host:

Hypervisor OS: ESXi 4.0 Releasebuild 294855
Systemboard: Intel S5000PSL
SCSI Controller: IBM ServeRaid 6M - 128 MB cache
RAID for Testing:  RAID5 built with 5 73GB 10K U320 disks.
RAID Stripe Size: 64KB

To do the testing I created two Windows 2003 guests with the following configurations:

OS: Windows 2003 Standard
Service Pack: SP2
Hotfixes:  All hotfixes presented over Windows update as of 11/15/2010
Virtual Disk: 12 GB dynamic
Virtual Memory: 512MB
NTFS Format: default (4K)

The only difference is that for SERVERB, the disk was aligned to 64K as described in the documents, and SERVERA had no alignment.

I then tested the servers using iozone.org version 3.53.  In order to isolate each server I rebooted the ESX host, then gave it 5 minutes to settle, and started up the Guest target alone with no other guests running with another 5 minutes of settle time before starting the testing.

The tests were done three times for each server and the results were averaged.  Then the results for each aggregate server were compared as a percentage difference.

Now the big question, is it worth it.  The short answer is yes,  I saw an improvement on average of about 10 to 40% per test.  There were points where the aligned disk was slower, but if you look at the aggregate of the data there is an improvement.  I need to do some more cleanup of the data but I will be posting it soon.

0 comments:

Post a Comment