VMM converged fabric v.s. MPIO

zfllj · Thu Jul 03, 2014 6:19 pm

What are the recommendations when dealing with converged fabric in Virtual Machine Manager 2012 R2?

Specifically, in VMM one can team NICs with say, switch independent / Dynamic hashing (to be able to use 2+ switches for HA) yet the documentation for SW VirtualSAN specifically recommends against NIC teaming for iSCSI traffic.

There are built-in port profiles in VMM made for iSCSI traffic shaping so it appears as if Microsoft is implicitly supporting a MPIO iSCSI scenario using virtual adapters but maybe this is not the case? Or is not supported by VSAN even if it is by MS?

Is a fallback scenario to leave individual physical NICs for iSCSI multipath and team the rest for the other networks (heartbeat, cluster, live migration, etc.) using virtual adapters?

It would be nice if I could leave my eight NICs teamed and just sprout a couple virtual adapters for multipath iSCSI traffic...

Fri Jul 04, 2014 11:14 am

With Windows Server 2012 R2 you actually *CAN* team connections and use that big fat pipe for iSCSI (of course if you use MSFT and not vendor-provided LACP stack). See:

http://blogs.technet.com/b/askpfeplat/a ... stion.aspx

However using MPIO is sill a preferred way to go with iSCSI in general and StarWind in particular (for back storage, for direct connections when running hyper converged it's a MUST as we
don't support direct memory access bypassing TCP stack with teamed stuff and will never support that scenario and for dedicated storage and compute layers on Hyper-V we prefer to use
SMB3 as a front-line connection protocol, iSCSI is pretty much history with MSFT).

zfllj · Fri Jul 04, 2014 4:52 pm

This may be answered in an upcoming white paper on best practices but:

Where does the Scale-Out-File-Server role reside in the picture?

For the Hyper-Converged model I assume they would just be VMs on two or more of the nodes but for the two tiered model (computer + storage) do you put the Hyper-V role on the storage nodes and spin-up SOFS VMs or do you spin them up on the compute nodes?

Fri Jul 04, 2014 9:06 pm

Yes, they are answered by whitepapers (converged scenario already published on the site so check for it) and dedicated setups.

SoFS role runs on a non-converged (dedicated) set of servers acting as a HA storage cluster.

No Hyper-V role on SoFS servers it's not supported by MSFT.

zfllj wrote:This may be answered in an upcoming white paper on best practices but:

Where does the Scale-Out-File-Server role reside in the picture?

For the Hyper-Converged model I assume they would just be VMs on two or more of the nodes but for the two tiered model (computer + storage) do you put the Hyper-V role on the storage nodes and spin-up SOFS VMs or do you spin them up on the compute nodes?