Mixing thick & thin disks on a single VM - benefits?

[MR-T]

I've been thinking about where to use thin disks and I already understand the pottential admin overhead of watching growth and how defrag could grow the disk to its full size but it got me thinking about the possibility of mixing thick and thin on the same machines.

Has anyone tried this or think it's a waste of time?

Because the Operating Sytem remains fairly static yet we seem to provision 40GB each time a new VM is created I think this would be the best candidate for thin. Essentially the OS requires half of what is provisioned and the remaining space is used as patches are applied. Although it's recommended to frequently defrag the hard disks, in reality this very rarely happens across an estate containing hundreds of severs so there's no chance of sudden disk inflation.

Because the applications and data are placed on a seperate disk inside each VM and the expected growth varies from box to box, although this is also a candidate for thin I was thinking thick might be better here. Because it removes the administrative overhead of constantly worrying about the day someone copies a 10GB database onto the thin disk and pushes the VMFS to 98%, this approach could work.

I estimate that by using this approach I would save 20GB per VM across a 300 VM estate which would be 6TB.

Please tell me if I'm totally bonkers.

[depping]

You are not bonkers and I guess it would be a good approach if and when defrags don't take place as then you will more than likely end up with all bloated VMs any way.

-d

[golddiggie]

Use of thick/thin provisioned vmdk's is (as far as I know) still one of VMware's "it depends" items... I would be ok with using thin provisioned for just the OS drives (aka C:). For the other drives, it depends on what they're being used for and if there's any benefits from going thin or thick. If it's a higher IOPs application installed on the drive, or it's being accessed heavily, then I would go thick. If it's a pretty light load, then thin could work out well too.

When I'm creating VM's, I review what it will be used for and then make the drives either thick or thin accordingly. This has worked out pretty well so far. The parameters I stated above are the general ones, with exceptions being made as necessary. Most of the time, if they are changed, it's a move from thin to thick for disk IO reasons. Of course, your stroage infrastructure will play a part in this. Really good/high performance SAN's can compensate for the performance hits you could see from thin provisioned vmdk's.

Some operating systems handle being on thin vmdk's better (or so it seems) than others. Linux, IMO, is a good candidate for thin provisioning. SQL or Exchange would be solid candidates for thick.

Just my experiences so far, and $0.25 worth... :smileygrin:

[AndreTheGiant]

I use this approach in most case.

For example the data disk for a file server usually is thin.

The only "issue" is that you cannot use storage migration or a Storage vMotion to decide the type of disk.

You can only convert all VM's vmdk to one format or another (or keep the existing types).

Andre