Has anyone adjusted the queue depth on their disks to help performance

[kenwalters]

We have a few AIX 5.3 machines attached to an IBM DS4300 SAN. Since another region is running the same applications on another machine attached to a Xiotech SAN and they were experiencing response issues and slow backup times, Xiotech recommended changing the queue-depth settings on the disk drives. Once this was done the performance increased dramatically. They went from a queue depth setting of 2 to 65.
Our corporate office wants us to try the same change, but I'm not sure we will reap any benefits from our IBM SAN. Has anyone else made these tweaks on a pSeries attached to an IBM SAN? If so, did it increase performance.

 

[itsp1965]

Hi Ken, have you tried the IBM Techdocs library. This resource has 100s of documents from IBM Techline support folks covering a variety of topics. Try a lookup of DS4000 (since it covers the entire product family) to see if you hit something. Otherwise you can always go to the IBM redbooks site which has technical guides covering their product lines.

 

[khalidaaa]

Why do you have to change the queue depth! As recommended by IBM, that has to be the last resort! Have you tried distributing the load on multiple disks? Have you tried adjusting the LTG size of the vg? Have you tried tuning the disks' pbuf?

http://www.ibm.com/developerworks/aix/library/au-aix5l-lvm.html

I believe the SAN itself has its own tuning as well which worth looking at as well!

Regards,
Khalid

 

[kenwalters]

Personally, I don't want to make the change, since our performance seems to be OK. Plus everything I've read states that there should be no need for any change with IBM disks.

 

[cryptoadm]

I've changed the queue depth before for NetApp disk and it made a positive difference in performance. This change was made based on NetApp documentation.

You can't determine whether it will be effective or not unless you have performance data before and after the change though. You have *some* basis that it does have a positive impact based on another region. Although they have a different SAN. Read the IBM disk documentation.

That is my opinion with 14 years of AIX experience.

 

[kenwalters]

Thank you everyone who responded. I did check the disk documentation and using the formula's described found there can be slight adjustments to the queue depth. I will get performance data first to really determine if there truly is a need for the change though.

 

[khalidaaa]

Could you please share us the formula's?

Thanks.

Regards,
Khalid

 

[kenwalters]

Here is the formula from the IBM DS4000 Best Practices and Performance Tuning Guide. I found it with the rest of the IBM Redbooks.

For firmware level 6.1x and higher: 2048 / (number-of-hosts * logical drives-per-host), and

for firmware level 5.3 and 5.4: 512 / (number-of-hosts * logical drives-per-host)

As an example: A storage server with 4 hosts with 12, 14, 16 and 32 logical drives

attached respectively would be calculated as follows:

2048 / 4 * 32 (largest number of logical drives per host) = 16

512 / 4 * 32 (largest number of logical drives per host) = 4

If configuring only at the HBA level, you can use the formula: 2048 / (total number-of-HBAs), and 512 / (total number-of-HBAs) for the respective firmware levels.

Important: Setting queue depth too high can result in loss of data and possible file

corruption; therefore, being conservative with these settings is better.