MPIO ALB Performance

[dataerror]

Hello,

During testing of my EVA4400, I noticed that when I enabled ALB in MPIO, it increased read performance significantly.

I am aware the ALB basically allows for the owning controller to handle all of the reads to the LUN.

In practice however, I have a SQL server that had a stuck job that was consistently performing reads. As a test, I enabled ALB and noticed that controller utilization went down, but so did the read perfrormance. Fom about 90 MB/s to 50MB/s.

Is this expected? As a 'best practice' should ALB be enabled?

[Peter Mattei]

What is the layout of your LUNS?
How many LUNs are assigned to the very SQL server?
If you have several LUNs all owned by one controller this could be the reason.
Try distributing the LUNs to both controllers.

Cheers
Pete

[dataerror]

Thanks for the reply.
I've got about 25 LUNs in total so far with more migrations on the way. Load for these LUNs is distributed amongst the 2 controllers, however, all LUNs belonging to a particular system are assigned to the same controller. In the case of the SQL server, it has 5 LUNs. The overall controller utilization is pretty low (<20%) and performance analysis of the server shows only 1 LUN to be active (the one that the job is reading from).

Would distributing the controller LUN ownership still matter much in my case?

Why does enabling the ALB have such a significant negative performance impact?

[Peter Mattei]

This is strange indeed.
If you are just reading from one LUN without other traffic going on I would expect better performance with ALB=on.

Can you verify that the IO is really going through the LUN owning controller only?

Try moving the LUN to the other controller anyway to see if the behaviour stays the same. If not it could be a SAN issue.

Pete