Why does ESXi NFS datastore outperform VM NFS mounts ?

[fletch2007]

We are seeing a 3x performance improvement when the VM is presented Netapp NFS storage as a guest OS disk vs the VM NFS mounting the same space.

This is problematic in the case where our apache VM pool needs to mount a shared web tree - performance suffers - the shared mounts can not be make "local".

the question is why the 3x performance difference?

Specs:

Dell 610s 196Gb Ram

vSphere 5.0 U1

Netapp plugin for vCenter provides best NFS settings

CentOS 5 VMs

VM NFS mount options: rw,noatime,bg,nfsvers=3,proto=tcp,hard,nointr,timeo=600,rsize=32768,wsize=32768

thanks

[fletch2007]

On a related note we had a Linux VM on netapp NFS datastore with IO performance issues.

Following VMware support's recommendation we were able to tune the Guest OS to improve performance:

http://www.vmadmin.info/2012/10/linux-vm-io-scheduler-tuning.html