However, you might be lucky and have a dedicated SAN for your SQL Server; maybe to support failover clustering perhaps. In this case it makes sense that you want to push the SAN to its limit to get the best I/O performance by whacking up the HBA Queue Depth to a much higher value. It’s a sound theory and you would assume that there is some sort of stupidly large value that might negatively affect performance. As it turns out, it’s true. However, after building a Windows Server 2008 failover cluster recently quicker than expected I had some time to test some different values and was surprised at the results which indicate that performance started to decline with an HBA Queue Depth of only 128.
I used SQLIOSim to test each configuration as I was interested in a general performance test to give me an idea of how SQL Server might perform, not a pure I/O throughput test. Here are some links on SQLIOSim you might find useful:
SQLIOSim available for download
How to use the SQLIOSim utility to simulate SQL Server activity on a disk subsystem
Understanding SQLIOSIM Output – This one is key to understanding the results!