Subject: Re: Realtime processes with sched_setscheduler Date: Fri, 20 Dec 2002 19:06:01 -0500 From: anton wilson <anton.wilson@camotion.com> To: Michael Svetlik <m.svetlik@ssi-schaefer-peem.com> On Monday 16 December 2002 04:36 am, you wrote: > On a common x86 PC, there' s a simple process, that lets the realtime > clock chip wake it up, each 500/2^n milliseconds. > Depending on a command line switch, it may incease its priority at > startup, using the well-known > > schp.sched_priority = prio; > err = sched_setscheduler( 0, policy, &schp); > > calls. > When the process is being run with the above priority control > (policy==SCHED_FIFO,prio==99), it shows timing deviations up to 40 ms, > compared to the predetermined desired value - I have one,or more, bad > processes running on my RedHat distro, that periodically think a lot ;-). > If being run with a simple 'nice --20', the jitter maximum is ONE > millisecond. > > The actual GNU libc manual says, that the POSIX-conform scheduler call > is just a fine-tuning of a process priority - the nice value is the > major parameter here. Does the kernel agree with that ? > > Michael Not sure what that question means. SCHED_FIFO, SCHED_OTHER, and SCHED_ROBIN processes are significantly different from each other in how the processor schedules them. You may want to read up on how the kernel scheduling policies work. I can't speak much for the niceness levels. But you probably should use sched_get_priority_max() to get the maximum priority. Maybe it's not setting it as high as you think. or you can use sched_getscheduler to see the difference in priority between using setscheduler and nice. -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
