Disk alignment on Windows system disk

[SorenHuus]

Hi,

Ive been reading http://www.vmware.com/pdf/esx3_partition_align.pdf and I wanted to make a template with the correct alignment of my system disk. I added an extra disk to an existing VM, followed to guide described, diskpart and format wise.

Then I created a new VM and added the newly created/partitioned/formatted disk and attempted an install. The Windows Server 2003 installation detects the file system just fine and installs on it. The problem comes when it needs to boot afterwards. It isnt able to. I checked and the disk is set active.

Is the alignment only for data/none-system disks?

- Huus

[Chris_S_UK]

What you have done sounds OK. What happens when you try to boot?

[SorenHuus]

I cant remember exactly. But I recall it being unable to read disk or something. Ill recreat the error and paste the exact message I get.

[jhanekom]

Check the boot.ini in the VM. Might be pointing to the wrong disk?

Also, disk alignment is usually specific to the storage vendor. It may not be necessary in your case. (I know I've seen EMC mentioning Clariion needs it, but that HP says EVA is fine without it.)

[SorenHuus]

Im using EVA. So I guess there is no need for it.

Thanks!

[jhanekom]

Just some more info on the matter: there is very little firm information on the topic of EVA disk alignment. I know some of the early Exchange 2003/HP EVA documentation explicitly stated that you should align EVA volumes.

However, I believe this was addressed in a subsequent XCS release. This document explicitly states that it is not necessary on an EVA 8000: http://h71028.www7.hp.com/ERC/downloads/4AA1-3935ENW.pdf. Not a definitive reference for the rest of the EVA range, I'm afraid.

Since it is definitely a controller software issue (the controller needs to understand the underlying data) and will primarily effect sequential writes on RAID5 volumes, I would suggest you test in your environment to make sure. (Sequential access to RAID1 volumes is not really affected, neither is sequential read from RAID5.)

Random IO tests will probably not yield much of a difference. Run multiple passes of sequential writes using something like IOMeter (with an outstanding queue length of at least 128) without disk alignment, then rinse & repeat with disk alignment.

The attached IOMeter test file should be a good start. You can get IOMeter here: http://www.iometer.org/