Disk Enclosures

EVA4400 and ESX 4.1 mru failover

 
SOLVED
Go to solution
inex
Super Advisor

EVA4400 and ESX 4.1 mru failover

I'm testing on a EVA4400 with ESX 4.1 mru failover settings. 4 luns, 2 on each controller. Luns are setup with path-A-Failover/Failback on controller 1, and Path-B-Failover/Failback on controller 2. During test i/o operations,I pulled out both host port cables on controller 1. All luns moved over to kcontroller 2 amd i/o continued. When I put back the cables in controller 1, all 4 luns remained on controller 2. I thought they automatically should have been moved to controller 1?

-inex
19 REPLIES 19
Uwe Zessin
Honored Contributor
Solution

Re: EVA4400 and ESX 4.1 mru failover

No, MRU means "Most Recently Used".

MRU is intended for active/passive arrays.

The EVA4400 is an (asymmetric) Active/Active array for which you should use RR (Round Robin).

RR (as well as MRU) are ALUA-aware (Asymmetric Logical Unit Access) which means they will talk to the controller which is owning the virtual disk.
.
inex
Super Advisor

Re: EVA4400 and ESX 4.1 mru failover

In my scenario will the lun fail back, if I used round robin?

-inex
Uwe Zessin
Honored Contributor

Re: EVA4400 and ESX 4.1 mru failover

How long did you pull the EVA controller host port cables?

The EVA does an implicit ownership transfer of a virtual disk if it detects that the majority of traffic goes through the other controller for about an hour.

With RR in ESX you should be able to force the ownership back to the intended controller using Command View-EVA. The ESX server should notice it after a short time and switch its active I/O paths to that controller.
.
inex
Super Advisor

Re: EVA4400 and ESX 4.1 mru failover

The cables was out for 2-3 minutes, an the luns changed ownership at once.

I wanted the luns to fail back automatically. Maybe fixed path is the only way to do this.

-inex


Uwe Zessin
Honored Contributor

Re: EVA4400 and ESX 4.1 mru failover

It's possible that the MRU failover code did trigger an ownership transfer - hard to say as not many people have access to these deep details.
.
Mark...
Honored Contributor

Re: EVA4400 and ESX 4.1 mru failover

Hi,
Under cv for each lun from the presentation tab you can set failover/failback (def is no preference) which should help.
Mark...
if you have nothing useful to say, say nothing...
Mark...
Honored Contributor

Re: EVA4400 and ESX 4.1 mru failover

Hi
Sorry ignore last!
Just read your post again and you have it already!!
Mark...
if you have nothing useful to say, say nothing...
inex
Super Advisor

Re: EVA4400 and ESX 4.1 mru failover

I have 10 ESX luns and I want 5 on each controller. In case of controller problems, I want the luns to failback to their original controller. MRU does not do this. Maybe fixed path will.

-inex
Amar_Joshi
Honored Contributor

Re: EVA4400 and ESX 4.1 mru failover

For the time being you can change the Vdisk settings to Path-A-FailoverOnly and that will bring back the LUN to controller A; then you can set it back to failover/failback again. I have seen this many times that EVA doesn't behave as expected for failover/failback settings and this is how I have been dealing with such situations.

Also, I would like to highlight that Round-Robin settings for ALUA compliance array (EVA4400) and OS (ESZX4.x) will allow hosts to send IOs to all ports (FP1 & FP2) of owning controller only. MRU settings will send the IO to only one port on the owning controller (either FP1 or FP2). You best choice for load-balancing is to use Round-Robin in conjunction with preferred Vdisk path settings.