Yesterday I received an email from someone. He wanted to know why he was limited to using either the “fixed” or “MRU” pathing policy for the LUNs attached to his MSCS cluster. In his environment they used round-robin for everything and not being able to configure all of them with the same policy was against their internal policy. The thing is that if round-robin would be used and the path would switch (by default every 1000 I/Os) the SCSI-2 reservation would need to be re-acquired on this LUN. (MSCS uses SCSI-2 reservations for their cluster devices) As you can imagine that could cause a lot of stress on your array and could lead to all sorts of problems. So please do not ignore this recommendation! Some extra details can be found in the following KB articles:
Matt Liebowitz says
Hey Duncan – one more KB link for you on this article:
Covers EMC PowerPath/VE and shows that if using a version later than 5.4 SP2 it supports MSCS shared RDMs.
I know you’re likely referring to the built in PSPs in vSphere and not third party MPPs, but figured I’d share anyway since it’s good information to have.
Cedric Blomart says
Concerning built-in PSPs, it is also to note that RDM luns used for MSCS are to be set with a “perennialy reserved” flag in ESXi 5.0.
If not, you might see your cluster nodes booting in hours (personal experience).
to continue on the kb hype:
hope it helps… i’ll get back to our storage team on powerpath/ve (a few month ago they were surprised by the licensing impact… we are then in the VMW_RR_PSP scenario).