in my epyc compute, when cpu go into very heavy workload, sometime farming timeout. so i increase farming pool size, but cpu io wait go high. i have a idea to spilt reading from farming, most of time only need little thread for get solution candiates, once got a candiate, i can use much bigger farming pool to calculate solution.
Not sure what you’re proposing exactly. Farming is basically all reading. And when doing proving all other operations are paused in farmer. If you have otehr stuff happening on the same machine you might want to tune process/thread priorities of those other workloads.