Default path selection policy for FC EVA8000 is MRU?

[MKguy]

We installed new vSphere Hosts and connected them to the FC SAN with an EVA8000. In ESX 3.x, the default path selection was fixed (as for active/active arrays) and we used to balance paths to the LUNs accordingly over the HBAs and EVA controllers.

Now with the new ESX4 host, the default policy is set to MRU (which was recommended for active/passive arrays in ESX 3.x) and ESX4 displays the array type as "VMW_SATP_ALUA".

Is MRU really intended and recommended in ESX4 for EVA storage? Should I set the policy to use fixed manually?

[marcelo_soares]

In the HCL is also as MRU only for ESX 4.0. The path policy at the HCL always is the one approved by VMWare and the SAN provider. You should not change it to Fixed - problems like path thrashing or slow response may occur.

Hope this helps.

[nielsvp]

Hello,

I've tested this a bit, and found that the default MRU settings does not give any benefits - worst case you end up using the non-owning controller for all traffic - later on you'll end up fighting path trashing.

Fixed and manual preferred path was the way to go in version 3.x - and the only way to balance paths.

Since vSphere is ALUA aware, and Round Robin isn't experimental anymore i would highly recommend taking a look at that setting instead.

From all my tests VMware does a really good job at selecting the optimized path (e.g. using all the paths to the controller owning to LUN)

Regards,

Niels Vejrup Pedersen

[srinivasvivek]

If you want to have both dynamic load balancing & availability of paths, try RR in vSphere.

For 3.5, this post should help you

http://virtualgeek.typepad.com/virtual_geek/2009/04/are-you-stuck-with-a-single-really-busy-array-po...

[supahted]

Did anybody actually tried the vSphere round robin path selection policy with a HP EVA?

blog:

[binoche]

you are correct,

MRU is default path selection for VMW_SATP_ALUA, and it will also select Active Optimized path as MRU;

FIXED is not recommended since it maybe select Active Un-Optimized path;

Round Robin will also select Active Optimized paths only by default but I do not know FC EVA8000 support Round Robin or not

binoche

VMware VCP

Cisco CCNA

[Paul11]

I was using Round Robin since ESX 3.5 (experimental support) with 2 EVA5000 without any problems for a very long time with about 30 TB of storage in a production environment.

With Vsphere there is also no problem in using RR on EVA's 5000 and 8x000 (Active/Active) in my environment. The only thing you had to know is, that you must set RR on every ESX-Host, because the setting is saved on the ESX-Server and not in the VM (when you are using RAW-Devices). Otherwise it get lost, when you move the VM to another host. The load balancing is excellent, it's exactly 50/50.

[sim411]

With Round Robin as PSP, did you specify in CV EVA a "prefered path/mode" in the advanced settings of your luns ? Or simply leaved it with "No preference" setting ?

[Paul11]

I have specified a prefered path for load balancing the LUN's between the 2 Controllers. Round Robin than makes the load balancing between the 2 Ports of each Controller.

[jpdicicco]

Hi Paul1,

Just trying to clarify your experience with RR on EVA8000: are all paths being used, or only the paths to the currently active controller?

Thanks,

JP

[supahted]

Hi jpdicicco,

Only the paths to the owning storage controller of the lun are used. So if you have a redundant setup with four paths, the I/O traffic is load balanced over the two paths to the owning storage controller. I wrote a blog article on this subject. If you want some more background information, check my article at

Regards,

Ted Steenvoorden

blog:

[Paul11]

Also your question is already answered I try to clarify the installation on my site:

There are 8 Front-Ports on the EVA8x00, so they are named FP1, FP2, FP3 and FP4 on Controller A and on Controller B.

We are Zoning FP1 and FP3 in Fabric1, and FP2 and FP4 in Fabric2

LUN1 is preferred to Controller A, LUN 2 is preferred on Controller B

ESX-Host1 is in a Zone with FP1 and FP2 (on each EVA-Controller, so you will see 4 way's to the disk, all are "active", 2 of them are "active (I/O)"

ESX-Host2 is in a Zone with FP3 and FP4 (of course one zone per fabric)

With Round Robin you can use now every port of the EVA8x00, if you have at least 2 ESX-Hosts, 2 LUN's and 2 HBA's in the hosts.

Maybe this will clarify the configuration.

[jpdicicco]

Thanks supahted and Paul1. I had read the docs, but wanted confirmation before I pursued using RR in my own environment. I appreciate you both explaining it clearly, providing additional references and (most of all) your anecdotal evidence.

JP