10Gb Network design HA for NFS

[jchilton]

Can anyone help me find a suitable ref architecture for high availability for NFS pse?

I've got a pair of 10Gb nics in an R730 VDI host and want to know best way to set up vmk ports, teaming, failover for a Dell EMC Unity300F array without single point of failure.

I've just read VMware Jan 18 white paper best NFS and none the wiser by Cormac. Not really relevant an dopens up more questions than answers.

[daphnissov]

Are you able to use NFS 4.1 in this environment?

[jchilton]

Think so. Am i right in thinking that v4.1 won't do DRS, SRM, vVols and a few other limitations.  At the moment we don't use those so i guess we are good to go.

I've just been reading about Dell EMC Unity Fail Safe Networking (FSN) which looks to provide another level of resilience in case of physical upstream switch failure.  Not sure if the ports on the Unity need to be in an LACP.

I'm thinking just add 2 x 10Gb vmnics (both active) on a vDS, each connected to a different pSwitch and teaming set to "Route based upon physical NIC Load"; only 1 vmkernel port for NFS traffic and let the hypervisor decide the path. I don't need any clever load sharing/balancing.  I'm guessing that if a pSwitch fails a route will be found via the other vmnic/pSwitch.

Is this a good assumption??

[daphnissov · Accepted solution]

Those features work with NFS 4.1 to my knowledge, certainly DRS does. If you can use NFS 4.1, then you can configure multi-path by setting multiple IPs on your NFS server and configuring ESXi with those IPs. It's important to note that VMware, in this nascent version of NFS 4.1, does not implement the pNFS standard, so true parallelization will not be possible, but you can "share" the load by using the above method. Otherwise, NFS configured for a vmkernel port will prefer a single physical NIC. The LBT method is more commonly used with VM traffic and not IP-based storage traffic.