Re: [patch] scheduler cache affinity improvement for 2.4 kernels

[Mike Kravetz <kravetz@us.ibm.com>]

On Thu, Nov 08, 2001 at 03:30:11PM +0100, Ingo Molnar wrote:
> 
> i've attached a patch that fixes a long-time performance problem in the
> Linux scheduler.

Just got back from holiday and saw this patch.  I like the idea
slowing down task dynamic priority modifications (the counter
field).  My only thought/concern would be in the case where a
task with maximum dynamic priority (counter value) decides to
use 'all' of its timeslice.  In such a case, the task can not
be preempted by another task (with the same static priority)
until its entire timeslice is expired.  In the current scheduler,
I believe the task can be preempted after 1 timer tick.  In
practice, this shouldn't be an issue.  However, it is something
we may want to think about.  One simple solution would be to
update a tasks dynamic priority (counter value) more frequently
it it is above its NICE_TO_TICKS value.

> (it would be nice if those people who suspect scalability problems in
> their workloads, could further test/verify the effects this patch.)

I'll try to run it on my 'CPU intensive' version of the TPC-H
behcnmark.

In addition, I have noted that this patch applies with minor
modification to our MultiQueue scheduler, and should be a win
in this environment also.

-- 
Mike