I had a question from a fellow blogger about the Fixed/Most Recently Used setting for a SAN’s path policy. This was related to an IBM SVC, which was only supported as an MRU setup at the moment, but as of ESX U3: IBM SAN Volume Controller — SVC is now supported with Fixed Multipathing Policy as well as MRU Multipathing Policy. (although the SAN guide still says it’s not…)
We can have a long discussion about this, but it’s plain and simple:
On an Active/Passive array you would need to set the path policy to “Most Recently Used“.
An Active/Active array must have the path policy set to “Fixed“.
Now I always wondered why there was a difference in these path policies. There probably are a couple of explanations but the most obvious one is:
MRU fails over to an alternative path when any of the following SCSI sense codes NOT_READY, ILLEGAL_REQUEST, NO_CONNECT and SP_HUNG are received. Keep in mind that MRU doesn’t failback.
For Active/Active SAN’s, the Fixed path policy a fail over only occurs when the SCSI sense code “NO_CONNECT” is received. When the path returns a fail back will occur.
As you can see, four against just one SCSI sense code. You can imagine what happens if you change MRU to Fixed when it’s not supported by the array. SCSI sense codes will be send out, but ESX isn’t expecting them and will not do a path fail over.
My deployment notes say:
Active/Passive – MRU only
Active/Active – Fixed or MRU (discrepancy here)
I can find the backup docs for this if we need to have a brawl on this one. All I really care about is the correct answer.
Signed,
Grasshopper
Oh.. it’s on!
the SAN Compatability guide can tell you which is actually supported, but the rule of thumb is
Just a question if it ever happened to you…using IBM DS4x00 SANs, you have to set a preferred path in IBM Storage Manager, let’s say you choose SP-A (Storage Processor A) for accessing LUN 1 and SP-B for accessing LUN 2 and you choose MRU as policy within Virtual Center (this is the default setup). When an host (blade in a BladeCenter) is rebooted, both paths become unavailable and the one which is restored first become active for VC. It could happen that is SP-B, so that creates a problem since both LUN (1 & 2) are accessed through SP-B, MRU doesn’t fail back and the SAN reports an error because LUN 1 isn’t anymore accessed through preferred path. That’s the situation I faced some times and I ended setting Fixed Path in VC to preserve preferred path setting in Storage Manager. How would you do in this case?
Thanks for feedback. (I hope I have been clear explaining it
Well, I have to say that we use the IBM SVC and we use Fixed pathing only…not using 3.5 u3 which supports fixed on SVC. Reason why we’ve done this since we got the SVC is because it is truly Active/Active. If you leave the default of MRU…good luck with all your LUNs on 1 port…depending on which storage adapter you decide to scan. This is because choosing a preferred node (think storage processor) doesn’t mean squat to ESX. This is because the preferred node setting in ESX is for SDD…IBM’s multipathing driver. Since ESX doesn’t have that driver, it knows nothing of a preferred node on the SVC.