Disk strategy inside VM - performance problem

[PiTRe44]

Hello,

we are in the process of benchmarking several design to implement vmware in some branch office. All the benchs use a single W2K VM. The VM files are all store in an internal RAID1 datastore. For the first bench, the VM was using a single vmdk file, with to logical disk C: and 😧 declared. To simplify backup, I have added a second vmdk file to store the 😧 drive.

But, doing some simple operations inside the VM (file copy, file extraction from C: to D:), I've noticed that using one vmdk is more efficient than using two vmdk (ie : extactring a 928 Mo .gz file from C: to 😧 with a single vmdk takes 2 minutes 30 seconds while it lasts 3 minutes 10 with two vmdk). When in production, this VM will perform nightly batches, so performances will be important.

Is this difference normal ? do someone have some experience about it ?

for information, the tests have been done using VI 3i and VI 3.5, on a DELL 2950 and a HP BL460 platform.

thanks in advance.

Pierre

By the way, excuse my poor english.

[spex]

Most time during copy is spend moving the head of your harddisk between the two areas of source and target.

I would guess, if you use only one vmdk you will have less distance between them.

Also the placement of your lun (inner or outer area) have impact on the speed.

Outer is fastest, since in one spindle turn more data move under the read/write head.

Regards Spex

[PiTRe44]

Thanks for your answer but I 'm still perplex ... We're using an internal Raid1 array and there's nothing else on the disks. So, i could suppose that the 2 vmdk are adjacent and that their placement on disk shouldn't have such impact.

But ok, I can't see a better explaination so ... I take it !

[pmuc]

extactring a 928 Mo .gz file from C: to 😧 with a single vmdk takes 2 minutes 30 seconds while it lasts 3 minutes 10 with two vmdk

Are you using "preallocated" VMDKs or "growable" ones?

With preallocated VMDK files (that are created with the full virtual HD partition size), I would expect less administrative file system overhead than with using two dynamic "growable" virtual disks. At least disk fragmentation should be lower.

[PiTRe44]

We are using preallocated disks only.

In the meantime, I do some additional tests, simulating our night batch. Although there's some disk interactions, it's essentialy data reading and processing with little writing at the end. And using this scenario, there is no performance difference between one or two vmdk files so we can go with this design.

thanks for your help.

Pierre

[oreeh]

FYI: this thread has been moved to the Performance Forum.

Oliver Reeh[/i]

[VMware Communities User Moderator|http://communities.vmware.com/docs/DOC-2444][/i]