Hi,
Page, Jeremy schrieb:
The only thing odd is that we did have ASIS run, I wonder if the fragmentation is from all the stuff getting de-duped (it's a bunch of VMs)
Then ASIS is the cause of the problem.
We have seen this kind of behaviour occur at several customers who are using dedupe and vmware.
The upside is: They get extremely good space savings The downside is: data locality issues
What basically happens is, the filer eliminates the redundant blocks, so ALL VMs on your volume will have to read many of their (deduplicated) blocks from the SAME DISK.
One could argue, the WAFL readcache would solve this problem, but currently it does not because the cache is not block aware (yet) - it is only file/block aware. That means even though 2 VMs share one common block on disk, WAFL cache sees it as 2 different blocks since they belong to two different files.
I have heard rumors this will be fixed some time in 7.3, but hey, that's just rumors.
Later, when the cache will be dedupe-aware, we will actually see performance improvements, since a block which has been read from disk once will be served from cache for several different files.
Solution to your problem:
Split up the VMs into 3 Volumes: 1. Just for swapping. No dedupe, no Snapshots 2. System Drive ("C"). Snapshots, Dedupe 3. Data Drive. Snapshots, no Dedupe
This way you will get the best performance and the best space-savings out of your Netapp Box and your VMware Infrastructure.
cheers, Olli
P.S. And don't forget about the partition alignment inside your .vmdk, that will also give a speed boost and at the same time reduce disk-load