We grew one of our Netapp VMware aggregates today from 46 to 67 spindles (300Gb 15K RPM disks).
Here is the before and after analysis quantifying the effect of the added spindles IOPs capacity from the VM's perspective:
http://www.vmadmin.info/2011/08/more-spindlesmore-vm-throughput.html
One caveat: to gain the benefit of the additional spindles, we cloned the VM out onto the expanded 67 disk aggregate to realize the throughput and latency improvement. Remains to be seen if WAFL will re-stripe the existing VMs over time or if some manual process will be required to get the full spindle count performance improvement.
You have a 10Gb connection to the NFS volume and could only achieve 95MB/s from one VM?
@Fred: Yes, this is further corroboration the IO is rate limited by the storage, not the connection to the storage
I might be able to push annother 10% with the PVSCSI driver - see:
http://www.vmadmin.info/2011/04/fusionio-esxi-pvscsi-vm-benchmarking.html
In this fusionIO benchmark we pushed 1300Mb/sec (> 16x the Netapp 67 disk aggr throughput)
So, on the re-striping topic, I've been directed to Netapp reallocation (its off by default):
"If you add disks to an aggregate, you can redistribute the data equally across all of the disks in the aggregate using the reallocate start -f command."
I've already used the reallocate measure -o /vol/<volumename> to determine ONTAP is recommending reallocation for some volumes
Will have to quantify the impact of volume reallocation from the VM's point of view as well!
thanks
