lkml.org 
[lkml]   [2006]   [Apr]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: RT task scheduling
From
Date
Hi Vernon,


On Fri, 2006-04-07 at 21:28 -0700, Vernon Mauery wrote:

> 1) Deterministic scheduling algorithms (SWSRPS). Basically, with uniprocessor
> systems (or smp with a global run queue), it was really easy to say, run the
> highest priority task in the queue. But when there are several queues that
> are independent of each other, it is difficult. According to SWSRPS, nr_cpus
> highest priority runnable tasks should _always_ be running (regardless of
> which queue they are on). This might mean that there are longer latencies a)
> to determine the nr_cpus highest priority tasks and b) because of cache
> issues.

Yep, and task cpu dancing. Everytime a High prio task preempts a lower
prio RT task, that RT task might be pushed to another CPU.

>
> 2) Maximum deterministic latency. A task should be able to say that if it
> relinquishes the processor for now, MAX_LATENCY nanoseconds (or ticks or
> whatever you want to measure time in) later, it will be back in time to meet
> a deadline.

Yep, but the more important thing than latency, is to make your
deadline. Sometimes people forget that and just concentrate on latency.
But that's another story.


>
> As I understand it, real time is all about determinism. But there are several
> places where we have to focus on determinism to make it all behave as it
> should.
>
> Priority A > B > C
> If a lower priority task C gets run just because it is the highest in that
> CPU's run queue while there is a higher priority task B is sleeping while A
> runs (on a 2 proc system), this is WRONG.

Argh, terminology is killing us all. For this to be wrong, B isn't
"sleeping" it's "waiting" while in the run state. "Sleeping" means that
it's not on the run queue and is just waiting for some event. Which
would be OK for C to run then. But if B is on the run queue and in the
the TASK_RUNNING state, it would be wrong for C to be running somewhere
where B could be running.

> But then again, we need to make
> sure that we can determine the maximum latency to preempt C to run B and try
> to minimize that.

And here I don't know of another way besides an IPI to preempt C. If C
is in userspace, how would you preempt C right a way if B suddenly wakes
up on the runqueue of A?

>
> Poof! More smoke in the air. I hope that clears it up.

It's as clear as my face was in High School ;)

-- Steve


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2006-04-08 06:53    [W:1.143 / U:0.052 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site