Yes, increasing the # of threads could be contributing to the longer processing time for concurrent operations. Increasing the -iqmt allows more 'stuff' to execute concurrently since more threads are created. However, those 'extra' threads are still running on the same 16 CPUs so there isn't any increase in the total compute power available. The slower execution time for operations running in parallel would be consistent with increased CPU contention.
What do you see reported for CPU utilization?
The recommendation from KBA 2936708 to increase the # of threads is to take advantage of the async IO by providing more threads for the IO operations. Since IO operations have a high latency compared to in memory operations, adding extra threads for IO won't introduce much extra load on the CPUs, rather it will reduce time spent waiting for IO operations.
Like all recommendations for default settings, the recommendation to set -iqmt to the max allowed by the OS is intended to be a good starting point. That doesn't mean that it will be the optimal value for all systems. As a tuning step you could try setting a -iqmt value halfway between your previous setting and the OC max allowed.
Bluesky
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.