What effect does HBA queue depth and max transfer size have on transmission overhead?

Question

This is a pretty straightforward question. What, if any, relationship do queue depth and max transfer size a QLogic HBA have on management overhead on a transmission? What I'm hoping to learn how (if at all) these settings would influence the amount of overhead "lost" to checksums and such (see Uwe Zessin's answer here)

Answer 1

There's a very small overhead on each frame. Max transfer size on the HBA being set too small can cost you a bit of overhead from sending more frames than needed, but the default (512KB) is larger than a standard FC frame (2148 bytes, 2112 bytes of it as payload). Each transfer from the HBA will be broken up into a sequence of frames anyways.

Your queue depth has no effect on transmission overhead, however it can have a great effect on your overall performance. If you have a larger lineup on your server than on your storage, you might see some application latency. Best practice is to try to line up your storage port's ability to queue IOs with your server's queue depth. If you have multiple servers on each storage port (who would do that?? :P), then you need to balance it as best you can. Each storage vendor will have a recommended queue depth for each OS, usually.