NPIVs and VMs

[Henry_He1]

Hi, I am looking to model workloads running on many VMs over an vSphere 5 / ESXi 5 host over Fibre Channel. I am not 100% certain if unique FCIDs and WWPNs are used for different VMs. Here's my understanding. If you know the answer, I'd appreciate it if you can clarify it for me.

My understanding:

- If a VM is simply using the Virtual Disk, then there is not a unique FCID / WWPN assigned to that VM.

- If a VM is configured to use FC NPIV (under VM Properties -> Edit virtual machine settings -> Options), then a unique FCID / WWPN is assigned to that VM. The host accomplishes that by using NPIV.

- Example 1: If a host has 40 VMs installed, and all of them only use Virtual Disk (i.e. no FC NPIV enabled), then the host will use one FCID (i.e. the HBA's N_Port FCID) and WWPN to send I/Os down the storage infrastructure for all VMs.

- Example 2: If a host has 40 VMs installed, and all of them has FC NPIV enabled, then the there will be 40 unique FCIDs / WWPNs observed coming out of the host's HBA.

Is the above correct? I would trace this myself if I have a FC analyzer, but I don't at this point. So I have to raise this question here.

Thanks

- Henry

[tomtom901 · Accepted solution]

Yes, you are correct! Spot on.

[Henry_He1]

Hi tomtom901, a follow-up question as I dig deeper into this. In the case of a single initiator (i.e. no NPIVs), how does ESXi unambiguously identify the SCSI/FC traffic from different VMs? Does it use an internal tracking system similar to NAT in IP, or does it make use of some existing SCSI/FC headers? Thanks!

[tomtom901]

Do you mean with the Fibre Channel LUNs as a datastore or as an RDM on the VM?

[Henry_He1]

I was referring to Fibre Channel LUNs as a datastore, but am curious about the behavior with RDM as well now that you mentioned it. How would ESXi differentiate SCSI/FC traffic for different VMs in both cases? Thanks.

[tomtom901]

Well, if you use Fibre Channel LUNs as a datastore in VMware, there is no reason to differentiate VM traffic because all VM's talk to the VMFS datastore in VMware. The way VMFS is built allows for a clustered file system solution that handels the VM disk writes and reads. Same as for iSCSI, VMFS is key in this. Whether it's Fibre Channel or iSCSI, VMFS assures all VM's can write simultaneously to the underlying storage.

With a RDM there is a one to one mapping of the FC LUN and the VM disk so I guess it uses some information stored in the protocol headers to differentiate traffic.