You would probably want to use your HBA (or vHBA) vendor's reference whitepapers for optimum performance. You can check if you are hitting your max DQLEN through esxtop; if you are not hitting the max, than adjusting this value won't make any difference. One word of caution, if you do increase your DQLEN, consider enabling SIOC, if you haven't already to ensure all of your VM's play nicely with each other.

hi

any idea what the queue depth should be? I see the default is 32 on esxi 5.1

should i increase it?

Hi,

the default Queue Depth depends on the used HBA's.

VMware KB: Changing the queue depth for QLogic, Emulex and Brocade HBAs

AFAIK, there's no similar KB article for Cisco CNA's.

As already mentioned by Paul, check with esxtop if you'll ever see QUED larger than 0 or ACTV constantly close to DQLEN.

When this isn't the case, there's no benefit of adjusting the queue depth.

Regards,

Ralf

For the longest time, Cisco hard-coded queue depth at 32 in the fnic driver. I had numerous conversations with project development managers and SE's regarding when it would be beneficial to increase this value. Their point was that all of their internal tests had proven to them that increasing it would not yield a significant increase in performance and could actually make performance worse. However, enough user demand was raised that they conceded and, as of fnic driver 1.5.0.45, end users can configure the queue depth and I/O throttle count of the CNA (VIC 1240 and 1280).

http://www.cisco.com/c/dam/en/us/solutions/collateral/data-center-virtualization/unified-computing/guide-c07-730811.pdf

The 1.5.0.45 fnic driver works with all Cisco vHBA's (Palo - M81KR, 1240, and 1280 VICs) and allows tunable parameters for each.

hi

this new driver is dependent on what kind of hardware I have on ucs right?  I still have b200m2s some with m81KR and 1280 VICs

some b200M3s

All are on vsphere 5.1U2

hi

i was able to change the queue depth to 128

but in esxtop

I still see some at 32

any idea?

I am running iometer tests and this is what esxtop shows. IT does not max to 128. Should i scale back on DQLEN?