Is NFS the way to virtualize?

I've been hearing more and more about how larger implementations should be NFS. Most of Netapp's virtual tools (RCU etc) are released first with NFS support and other protocols are added on after the fact. So should you be doing NFS?

I've talked before on this blog about how network bandwidth isn't really a constraint as you won't do anywhere close to 1Gb/sec of storage IO on a single ESX server, but you might at the Netapp, so make sure you're doing a multi vif (more than one nic active at a time).

Also, the move should now be towards putting multiple VMs in a single datastore so that you can get the benefits of ASIS or dedupe. I'm really coming around on dedupe and see it work quite well on filers, especially with VMware.

In order to get the best density of VMs per datastore, you should mount the datastore as NFS. NFS will allow you to easily resize datastores. In the latest vSphere storage best practice guide from Netapp they are recommending your FC/iSCSI datastores be between 300 and 700GB, which allows maybe 5-15 virtual machines. NFS datastores will scale much higher and will therefore give you better space savings and more virtual machines per filer.

The key to all of this is that you don't require a fibre channel infrastructure, which has been the case for a while. However, it is now starting to become best practice to not use it!

Cheers,

Chris