Re: Can't sleep less than 20 ms
Dong Liu (dongliu@lucent.com)
Sun, 04 Jul 1999 23:12:32 -0700
Rogier Wolff wrote:
>
> Dong Liu wrote:
> > Boussard Per wrote:
> > >
> > > On Fri, 2 Jul 1999, Andrea Arcangeli wrote:
> > >
> > > >
> > > > Set the policy of the scheduler as RT.
> > > >
> > >
> > > This will only work for root, no? The problem is the tv_sec||tv_nsec
> > > expression in sys_nanosleep(). I think this must go.
> >
> > This seems to be correct, I changed
> >
> > expire = timespec_to_jiffies(&t) + (t.tv_sec || t.tv_nsec);
> > to
> > expire = timespec_to_jiffies(&t);
> >
> > in sys_nanosleep(), now I got
> >
> > usleep ( <=10*1000) sleeps for 10 ms
> > usleep ( >10*1000 <= 20*1000) sleeps for 20 ms
> > usleep ( >20*1000 <= 30*1000) sleeps for 30 ms
>
> This may be a nice hack for YOU, but in no way should this be in the
> standard kernel.
>
> This looks fine:
>
> Clock interrupts + + + + +
> App is waken up: * * * *
> app calls usleep (9000) ^ ^ ^ ^
>
> but... if the application uses 8ms of CPU time before calling
> usleep again:
>
> Clock interrupts + + + + +
> App is waken up: * * * *
> app calls usleep (9000) ^ ^ ^ ^
>
> it will only sleep for 2ms before getting woken. And the manual states
> that the sleep will be -=* at least *=- the number of microseconds
> requested, and that you should always be prepared to be woken a little
> later due to scheduling overhead or other tasks running.
>
I already have a solution for this, I'll have one thread sleep for
exact 10ms and then post an event, another thread will do the
processing (I'm writing a RTP library, I have to send/recieve RTP
packets in the right interval) then wait for the event. In this
way, the processing thread will wake up at the moment of
(10ms - the time used to do processing).
Increasing HZ maybe another solution. but I don't the performance
implication.
Dong
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/