HPE EVA Storage

linux multipath + EVA 8100

 
SOLVED
Go to solution
Joshua Baldock
Frequent Advisor

linux multipath + EVA 8100

Hi All,

We have approx 35 RHEL5 servers that are connected to our EVA8100 SAN. We have configured multipathd on the servers for "round-robin 0" however when looking at evaperf we can see that these hosts are only using the path to controller B. These servers are not sending requests to Controller A.

We have confirmed that all zoning, HBA and CV configuration is correct. The servers can see all paths and all paths are active.

Does anyone have a White Paper of how to configure Linux hosts for EVA8100? Or any insight into why our Linux servers are only using one path, when configured for "round-robin 0"
8 REPLIES 8
Johan Guldmyr
Honored Contributor
Solution

Re: linux multipath + EVA 8100

Hi,

this might actually be "working as intended". Or that the hosts have ALB activated (can sense which controller owns the lun).

On the EVA, only one controller actually owns the LUN.

This means that if your host try to access the lun via the not-owning controller, the not-owning controller will pass the request on to the owning controller anyway (via mirror ports). Because of this, it is recommended to split the LUNs between the controllers (based on load preferably).

So Joshua, can you check on the vdisks? Is each vdisk's path set to Controller B?
Joshua Baldock
Frequent Advisor

Re: linux multipath + EVA 8100

Hi Johan,

Thanks for your reply.

I can confirm that the owning controller for these luns is "Controller B".

My understanding is that owning controllers on the EVA only affects writes? These hosts are also only using that path for reads.

So even if the hosts are actively only writing to the owning controller, why won't they read from both, as read requests can be serviced by either Controller?
Johan Guldmyr
Honored Contributor

Re: linux multipath + EVA 8100

Hi,

no, both reads and writes goes through the owning controller. Read req comes to the not-owning, it passes to the owning, this gives the answer back to the non-owning who gives it to the requester.

Write performance is the same even if you send it via the non-owning, because the mirrored cache needs to be updated.
Joshua Baldock
Frequent Advisor

Re: linux multipath + EVA 8100

Hi Johan,

Thank you for your replies.

Im wondering if you have a link to a user guide or white paper that clearly defines the owning controller functions for both reads and writes of the EVA?

As i did the HP EVA Course, it was a little while ago, and I believe our instructor advised that read requests were not concerned with the owning controller as both controllers have read access to all LUNs.

And so im just looking for a definite documented answer, to curb my current confusion.

Thanks again.
Johan Guldmyr
Honored Contributor

Re: linux multipath + EVA 8100

I'm sorry no I do not have it :/

Don't you get the material when taking the course?
Sivaramakrishna_1
Honored Contributor

Re: linux multipath + EVA 8100

Apart from Multipath policies, it is recommanded to set the LUN Load balancing across the controllers. You can set the 50% of LUNs Preferred controller to A and 50% to Controller B bases on the IO on the volumes.

Johan Guldmyr
Honored Contributor

Re: linux multipath + EVA 8100

Is that really recommended?

I was under the impression that instead LUNs should be divided between the controllers and then that hosts only access the LUN via the owning controller.

Because if a host access a lun via the non-owning controller the non-owning will just pass the request on to the owning anyway..
Uwe Zessin
Honored Contributor

Re: linux multipath + EVA 8100

That is what Sivaramakrishna meant with "LUN Load balancing".

You check how many I/Os go to individual virtual disks (call them "LUNs", if you like...) and then do a manual load balancing by assigning controller ownership via CV-EVA. Finally, configure your hosts to honor ALUA hints.
.