EVA 8100 with AIX and HP MPIO V1.0.1.0

[lin.chen]

Dear all,
EVA 8100 with 72 146G disks, connect to two switchs and then to IBM AIX systems using HP MPIO V1.0.1.0.
Performace is very slow. Check EVA data and found all the I/O only go through one controller.
How can I solve the problem?
Could you give me some advice how to config load balancing on MPIO?

Thanks and regards,Louis

[Tom O'Toole]

Check out this program that gets installed with MPIO:

/opt/hphsv/lbhsv

It can write a script with commands to help you balance your LUNS across paths on each system.
You could also set preferred controllers on the EVA for each vdisk.

It's not clear this is your main problem however, there could be a number of other factors.

[Ivan Ferreira]

>>> all the I/O only go through one controller.

This may not be a problem, it depends of the controller utilization, if the utilization is low, then this will not be a performance problem.

In fact, if the controller utilization is low, and all LUNs are owned/served by this controller, this is a correct configuration.

[Tom O'Toole]

The trouble with AIX MPIO is that, left to its own devices (badabump!), it sets the active path for all LUNS to be path id 0 (which ends up being the same fabric path for each hdisk device). Path priorities must be modded to cause it to use other paths.

I agree that there's a good chance this isn't his problem, and more info should be provided (like specifically how very slow performance is, for example).

[Ed Barnum]

First check the path priority with an "lshsv -p"; they are probably set to a priority of "1". To lower the priority of a secondary path, you can use a command like:
# chpath -l hdisk2 -p fscsi1 -w "50001fe1500adc8f,4000000000000" -a priority=2

The device must not be busy to change the path priority (volume groups that own that hdisk should be varyed off or exported).

Should you decide to use the "lbhsv" command Tom mentioned, I would recommend doing a:
# lbhsv -e ALL >/tmp/enable_static_load_balancing.sh
# cat /tmp/enable_static_load_balancing.sh

One other thing to check is queue depth on the hdisks with an "lsattr -El hdiskn". If you are running EVA firmware XCS 6.x then the queue depth can be greater than "1".

[Alexp_1]

tirebiter, thank you for the info you posted in this thread.

I know this thread is 2 years old now, but it's exactly what I was looking for.

You've just made my life a lot easier.

Cheers, Alex.