Dustin Marquess<dmarquess%gmail.com@localhost> wrote:
I'm curious on the status of the M2 scheduler (SCHED_M2) in -current.
Is there any reason not to use it, any known gotchas?
5.0 and later are already using, for a long time by now, the essence of
what was originally SCHED_M2, which is a dispatcher, more specifically,
per-CPU run-queues and balancing (SMP scalability is in these). Hence,
SCHED_M2 option these days is merely a simplistic SVR4-style scheduling
algorithm for time quantum and priority calculations.
I kept it mainly for experimental purposes, but for some workloads it
might have better behaviour, e.g. higher interactivity. Also, as Lars
noted, priority recalculation is O(1) for M2. So, test and see which
suits your workload better.