Dear all,

For some reason, I want to have 10 processes share the same CPU core by affinity.

As you know, to achieve CPU sharing between processes, we can exploit the linux round robin (RR) scheduling policy by calling sched_setscheduler() to set the process's scheduling property.

For linux kernel 3.9 and later, we can set the RR scheduling interval by changing the file "/proc/sys/kernle/sched_rr_timeslice_ms". The default RR scheduling interval is 100ms. That is, all the 10 processes can be executed once every second.

My questions are

1) I want the RR scheduling interval to be 100us, i.e., each process is executed once every 1ms. But I failed to set the "/proc/sys/kernle/sched_rr_timeslice_ms" with a value of less than 1ms. I want to confirm if this is the truth. That is, the minimum RR scheduling interval supported by current linux kernel is 1ms.

2) Is anybody know what minimum scheduling interval is supported by Windriver linux? As Windriver linux is widely used for embeded system which requires good OS real-time performance.

3) I wonder if I can have all the 10 processes to be executed at least once within 1ms by using mutex. That is, one process executes for 100us, then unlock the mutex so that another process can lock and execute. Does this mechanism work?

4) If the above mentioned method doesn't work, what if the 10 processes are changed to 10 threads. Is there any high efficiency thread switch mechanism avaliable?

5) Does the function set_scheduler() only apply to process, but not thread?


Any responses will be highly appreciated.

Cheng

The lower limit of the scheduling interval is dependent on CPU speed. As you decrease the scheduling interval you increase the percentage of CPU time used to schedule. At some point decreasing the scheduling interval causes response time to deteriorate because too much time CPU time is being used by the scheduler.

No. That method is counter productive. When a process is being executed it may go into a wait for I/O, timer, etc. Then the scheduler turns the CPU over to the next process in the scheduler queue which is ready to run. If all of the other processes in the scheduler queue are waiting on a mutex then none can be run if the current process goes into a wait.


It still won't work.

I don't know. After reading the set_scheduler() man page I think that the answer is yes.

http://linux.die.net/man/2/sched_setscheduler

-----------------------
Steve Stites

1 members found this post helpful.