EVA on hp-ux 11.31

[David Dilly]

Is it true that,
Half of the paths for an eva is shown STANDBY and the other is shown ACTIF ?

Thanks

[Olivier Masse]

That's what I have here.

[David Dilly]

So it's true ?

[Olivier Masse]

Yes. I wouldn't worry about it, I have the same with an EVA6000. I have two HBAs & four paths, and it activates one path on each HBA. On systems hooked up to an EVA8000 which have 8 paths, it activates only 4 of the 8 paths.

I don't know the exact algorithm that decides which and how much paths are activated but I did some resiliency tests last year by unplugging HBAs and it failed over correctly.

Bonne chance

Olivier

[Gordon Sjodin]

I think it is because the controllers are really Active/Passive - roughly meaning that it is less overhead for all trafic to stay on one controller. -- Just my thoughts.

[Duncan Edmonstone · Accepted solution]

David,

Yes thats correct - the EVA is a ALUA(asymmetric logical unit access) array - the same as most mid-range arrays (ceratinly the EMC Clariion is as well and I think maybe the HDS midrange array too).

This means for any given LUN. the IOs are handled by one controller - IOs can be received by the other controller, but it just acts as a proxy and passes the IO to the owning controller. Usually on a ALUA disk array LUNs are owned by alternating controllers (i.e. LUN 1 is owned by controller A, LUN 2 by controller B, LUN 3 by controller A etc.)

Some MPIO subsystems in some operating systems will send IOs to the non-owning controller, some won't... in the case of the MPIO subsystem in the 11.31 storage stack, it only sends IOs to the owning controller unless it fails...

HTH

Duncan

[Gordon Sjodin]

You may notice with an "iostat" commmand if you have autopath installed that the traffic is distributed quite evenly, in my case, over 4 of the 8 paths.

[Duncan Edmonstone]

Gordon,

This gets confusing... strictly speaking when an array is "active/passive" _all_ the LUNs are serviced by just one controller and IOs to the other controller will be rejected. On the EVA, both controllers are active, but for an individual LUN only one controller actually does IO.

HTH

Duncan

[Gordon Sjodin]

I agree it is confussing from the perspective that an EVA is an Active/Active SAN but without the full Active/Active capabilities hense leaving it definable as Active/Passive -- that is my take.

[Olivier Masse]

Duncan, thanks for the explanation, you made my day. :)