preemptable, priority based scheduling is your solution. if you are in a latency sensitive environment, the "fairness" or "throughput" of the scheduler are not

preemptable, priority based scheduling is your solution. if you are in a latency sensitive environment, the "fairness" or "throughput" of the scheduler are not things you care about. next most important is 'pinning" threads to cpu cores, such that cache misses (or faraway memory) are minimized.