I've been working with this today as well. HP just finished our 7200 install and we've been moving VMs over from an EVA4400 for the last couple of days. One VM with a very large datastore used for backing up remote sites had grown pretty large, but we were able to delete about 500GB off of it before migrating to the 7200. The volume on the datastore was thin-provisioned. I thought the Storage vMotion would clear those blocks out, but I forgot I needed to run sdelete on the guest first. I did that and then ran the unmap commands, but that still didn't do it. I know I could have Storage vMotion'd it again, but that would take several hours due to the size. I finally found this command in a VMware article:
vmkfstools -K (volume.vmdk)
Shut-down the guest in question. You go to a command prompt on a host that has access to the datastore. Find the .vmdk file you need to shrink and then run that command against it. It calls it "punching zeroes". I guess it is an in-place way to do the same thing as Storage vMotion'ing a thin volume to clear space.
Once I did that, I ran the "esxcli storage vmfs unmap" and vSphere now shows the free space I was expecting. The showvv -s command on the 7200 shows the space still in Reserved, but User is now down where I expect it should be. Hopefully some process will free that Reserved space back up at some point.
Just wanted to give my observations on this same issue.
|