We're about to upgrade our virtual environments to vsphere 5 and i'm going back and forth on how to configure NMP on my esxi hosts. We don't own Powerpath /ve so we have to use native. Our hosts are connected to an EMC VNX7500. From the folks i've spoken with, it seems my best options are either FIXED or MRU. Does anyone have an opinion on which to go with based on our array?
Any insight is much appreciated.
The VNX should be switched to failover mode 4 ALUA, see step 7 page 44, failover 4 mode ALUA is required for VAAI to be enabled.
ALUA PSP setting should be set to fixed or Round Robin.
See page 79 of VNX port configuration
Although Fixed Path is the preferred PSP for VNX, Round Robin is also of interest in some environments due to the ability to actively distribute host I/O across all paths.
I usually use Round robin so I can actively distribute host I/O across all paths but you should read the caveats and decided which one will suit you better.
The link above is to the VNX on vSphere documentation released this year.
We have a VNX7500 and our storage team told us to use round robin...we were promptly flooded with tresspassing errors. We are in the process of switching everything back to FIXED. Also, as far as I know, round robin doesn't support actually using more than 1 path at a time so there's no real performance benefit I can see unless you buy Power Path.
In fairness to round robin, we might be getting so many tresspassing errors because we have multiple storage groups accessing the same LUNs...poor design on our part
The EMC documentation says tresspassing is an issue with Round robin, so stick with FIXED unless you have a business need for RR...in my opinion.
I am referring to VMware's definitions of active/active vs active/passive:
From "VMware SAN Conceptual and Design Basics"
In this definition, a VNX, with or without ALUA 4, is an Active/Passive array as trespasses are needed in order for each SP to service the LUN.
Under this definition, a VNX is active active. Any given LUN can be serviced through either SP without significant performance degredation. Sending an IO to the non-preferred controller results in a performance penalty of less than 1ms, on average.
Trespasses have not been required since the VNX was released 2 years ago.
Has anyone got links to the latest doco pointing to this.
Sent on iPhone please forgive spelling mistakes and brief messages.
For more detail on the ins and outs see this blog entry