Re: 2.1.126 SMP scheduling

Scott McNab (jedi@tartarus.uwa.edu.au)
Wed, 28 Oct 1998 23:51:16 +0800 (WST)

> On 28 Oct 1998, Andi Kleen wrote:
> > In article <36360441.BCC187F2@ecn.purdue.edu>,
> > Noah Beck <noah@ecn.purdue.edu> writes:
> > > If, for example, I run four parallel loops as below, the interactive
> > > remote performance of the system is something like, if I hold down
> > > the <return> key so tcsh prints out a bunch of prompts, I only get
> > > about one or two prompts printed per second on average. If one were
> > > to move the mouse on the X console during this time, it would be
> > > extremely jumpy. Attempting to interactively edit files is an exercise
> > > in cursor prediction, as well.
> >
> > I see similar symptoms on both SMP and UP. If I nice the cpu hogs it
> > gets better, but per default the scheduler is very nasty to
> > interactice processes. This seems to be a regression compared to 2.0.
>
> Maybe my scheduling patch can help with this. One of the
> things the scheduling bigpatch does is heavily favour
> interactive processes and at the same time slightly lower
> scheduling overhead.
>
> If my patch helps, we could integrate part of it in the
> kernel. The entire patch consists of very trivial things
> and is certified to work stably...
>
> You can get the patch from my home page.

I have just applied the patch to 2.1.127pre2 and it has improved
responsiveness of the system considerably when there are a number of
CPU bound processes running.

I am not familiar with the SCHED_IDLE components but I imagine the other
parts of the patch could be merged into the kernel without much debate.

Scott

-
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/