How best to balance load between EVA controllers?

[Jeremy C]

EVA 8100 w/ CV EVA 9.2

"evaperf cs" shows controller B as always having significantly more CPU% and Data% as controller A (i.e. 95% vs 5%).

"evaperf vd" shows that nearly all vdisks are "online to" controller B.

Should I start specifying a preferred path/mode via CV EVA to push some onto controller A with "Path A-Failover/failback"?

[Uwe Zessin · Accepted solution]

You can try that, but not all multipath handlers in all operating systems recognize this preference properly. Often it is necessary to make adjustments on the server, too.

Can you tell what OS and version you are using?

[Jeremy C]

Basically two types of servers here:
1. Windows 2003/2008 with MPIO recently changed from ALB=N to ALB=Y (that should help).
2. ESX 3.x

After some additional reading on the forums, I've come to the following conclusions. Please correct me if I'm mistaken.
1. (Windows - MPIO w/ ALB=Y)
Set preferred path on vdisks to evenly divide the work. Or would they get divided over time automatically by the EVA?

2. ESX 3.x
Set preferred path on vdisks to evenly divide the work... AND ... make sure that the ESX preferred paths match with the preferred path setting in CV EVA. If it does not match then the EVA will be getting hit on the proxy controller and switch the vdisks back to the other controller.

Another question: I've read that if >60% of the requests for a vdisk are hitting the proxy controller then the EVA will switch ownership of that vdisk...but will it ever change ownership of vdisks because one controller is under heavier load than the other?

[Uwe Zessin]

Yes, ALB=Y is the way to go on Windows - plus a recent MPIO. I've seen that in one version ALB was not honored when one out of three FC adapter ports did not 'see' the EVA, because the customer used a dedicated backup fabric.

You can assign them manually or use no-preferrence. In that case the EVA will do round-robin assign vdisk ownerships during boot.
If you know the load, I think it is better if you distribute them across both controllers.


On ESX 3 you must configure fixed path on all hosts and make sure all ESX hosts hit the same controller - otherwise the first path detected is chosed for I/O.
As you already know: if the EVA notices that the majority of traffic goes through the proxy controller it does an ownership switch.
If the ESX 3 setup is not correct, you can cause permanent vdisk ownership switches, which are not good.


I got told about claims that the controller did some "load balancing" (when I was troubleshouting and found the MPIO problem I mentioned above).
I've never heard them myself, I've never seen that behavior on any EVA, I beleive that claims about EVA CPU or traffic load balancing are false.

[Jeremy C]

Thanks so much for the answers!

One clarification:
"the EVA will do round-robin assign vdisk ownerships during boot" ... You're referring to the boot up of the server?

[Uwe Zessin]

No, I meant when the EVA storage controllers boot, it looks at all the virtual disks preferrence setting and if there are virtual disks with no-pref it will assign ownership of the first vdisk to controller-A, the second vdisk to controller-B, the third vdisk again to A and so on.