I seems to unable to find a good reference to find the best number of thread_pool_size.
I have a very read heavy db server (barely any write). percona 5.7 on 16 VCPU + 32GB RAM.
i was using 80 thread_pool_size. When i go to 16-32 thread_pool_size i am seeing the latency gets worse. But latency gets better with setting the stall limit from 500 to 100.
I am trying to benchmark different TPS using sysbench but they mostly shows similar result (4K clients) between using 16 32 64 96 128 TPS (default 500ms stall limit).
What would be the best way to find the right size for TPS? (NCPU <= TPX <= NCPU + NCPU/2) doesnt seem to work best here.