I was wondering if both scheduling (contention) scopes (SYSTEM and PROCESS) are used in the Linux kernel scheduling logic. POSIX supports PROCESS scope pthreads (MXN model) and SYSTEM scope pthreads (1X1 model) for user level and system visible schedulable units, resp. Article "The Native POSIX Thread Library for Linux", by Ulrich Drepper and Ingo Molnar (Red Hat) suggesting that PROCESS contention scopes were overkill and thus discouraged.
Does kernel 3.x and 4.x support PROCESS scope Pthreads as well or just SYSTEM scope?
Also, what is the scheduler now used for SCHED_OTHER class?
Has anyone done performance or behavior analysis for SCHED_OTHER, FIFO and RR? These have been around for quite some time now. I believe that different classes of workloads (eg HPC or RT) could be positively or negatively affected by the choice of scheduler. But are there any concrete performance results ?
Thank you! --Michael