2.5.74-mm1 and Con Kolivas' CPU scheduler work

2.5.74-mm1 and Con Kolivas' CPU scheduler work

[ismail (cartman) donmez]

Hi,

With Con Kolivas' cpu patch I got many slow downs. When I try to run 
CodeWeaver's CrossOver office box gets unresponsive at least for a one 
minute. While starting KDE ( when splash screen is running ) mouse gets too 
sluggish. 

I tried Con's patch with 2.5.73-mm3 too and had same overall bad affects. 
2.5.73-mm3 works very good.

Anyone has an idea whats might be wrong?

Regards,
/ismail

-- 
An Eye for an Eye will make the whole world blind - Gandhi 

Re: 2.5.74-mm1 and Con Kolivas' CPU scheduler work

[Con Kolivas]

On Fri, 4 Jul 2003 14:11, Tom Sightler wrote:
> On Thu, 2003-07-03 at 22:29, Con Kolivas wrote:
> > Ah yes I believe I know this issue. The problem is the parent spinning
> > madly waiting for the child and it is the parent that starves the child.
> > While this is not a fix, if you can reproduce the problem can you try
> > changing CHILD_PENALTY in kernel/sched.c from 50 to 100 and see if that
> > makes the problem go away? I mentioned this hidden in a thread a while
> > ago, and am trying to get a reasonable fix.
>
> Well, perhaps I spoke too soon about this particular issues.  I have
> just compiled 2.5.74-mm1 and it seems to be much better behaved that my
> previous kernel (2.5.72-mm2).  I can no longer reproduce the issue with
> this new kernel but the problem is easily reproducible with my older
> kernel.  Does 2.5.74-mm1 have a recent version of your patches (I know
> it has some variation of your patches but you've been cracking them out
> pretty quick lately).  I've run testing with my horror cases of
> Crossover Plugin and multiple Crossover Office applications running
> simultaneously and all programs seem responsive, these cases caused all
> kinds of audio skipping and pauses on the system before.
>
> I'm still running some other tests that seem to be showing some
> strangeness but I need to do some more test on both kernels before I
> really reports them.
>
> Thanks for your hard work on these issues, for my workload things seem
> to be getting quite a bit better.

Heh well that's nice. The last patch posted on my site ( 
kernel.kolivas.org/2.5 ) is the same as in -mm1. There is no newer patch yet, 
but I've got some changes coming to apply on top pretty soon. Hopefully these 
will address your "strangeness". :)

Con

Re: 2.5.74-mm1 and Con Kolivas' CPU scheduler work

[Tom Sightler]

On Thu, 2003-07-03 at 22:29, Con Kolivas wrote:
> Ah yes I believe I know this issue. The problem is the parent spinning madly 
> waiting for the child and it is the parent that starves the child. While this 
> is not a fix, if you can reproduce the problem can you try changing 
> CHILD_PENALTY in kernel/sched.c from 50 to 100 and see if that makes the 
> problem go away? I mentioned this hidden in a thread a while ago, and am 
> trying to get a reasonable fix.

Well, perhaps I spoke too soon about this particular issues.  I have
just compiled 2.5.74-mm1 and it seems to be much better behaved that my
previous kernel (2.5.72-mm2).  I can no longer reproduce the issue with
this new kernel but the problem is easily reproducible with my older
kernel.  Does 2.5.74-mm1 have a recent version of your patches (I know
it has some variation of your patches but you've been cracking them out
pretty quick lately).  I've run testing with my horror cases of
Crossover Plugin and multiple Crossover Office applications running
simultaneously and all programs seem responsive, these cases caused all
kinds of audio skipping and pauses on the system before.

I'm still running some other tests that seem to be showing some
strangeness but I need to do some more test on both kernels before I
really reports them.

Thanks for your hard work on these issues, for my workload things seem
to be getting quite a bit better.

Later,
Tom

Re: 2.5.74-mm1 and Con Kolivas' CPU scheduler work

[Con Kolivas]

On Fri, 4 Jul 2003 12:09, Tom Sightler wrote:
> On Thu, 2003-07-03 at 19:49, Con Kolivas wrote:
> > On Fri, 4 Jul 2003 02:36, ismail (cartman) donmez wrote:
> > > Hi,
> > >
> > > With Con Kolivas' cpu patch I got many slow downs. When I try to run
> > > CodeWeaver's CrossOver office box gets unresponsive at least for a one
> > > minute. While starting KDE ( when splash screen is running ) mouse gets
> > > too sluggish.
> >
> > While your wm starting if your mouse is sluggish it is probably not a big
> > problem, and it is part of something I'm already working on. The
> > codeweavers problem is impressive though; I don't have codeweavers and
> > have not seen anything like it. To get a full picture it would be helpful
> > if you could run top as root say reniced to -13 just so it doesn't miss
> > anything, and then start codeweavers, watch top and tell me what is
> > spinning away hogging the cpu.
>
> My guess is that he's seeing the same issue that I reported a few weeks
> ago with Crossover plugin.  That report generated a fairly long thread
> and a lot of patches and testing, but little actual progress.  Basically
> wine seems to have multiple threads and a client/server architecture and
> seems constantly spin on a pipe passing info back and forth to each
> other.  It seems one thread gets all the attention and starves the
> others although the others are critical for the main process to actually
> make progress.
>
> I've been wanting to do some testing with your patches so I'll give them
> a spin tonight and see how this acts.

Ah yes I believe I know this issue. The problem is the parent spinning madly 
waiting for the child and it is the parent that starves the child. While this 
is not a fix, if you can reproduce the problem can you try changing 
CHILD_PENALTY in kernel/sched.c from 50 to 100 and see if that makes the 
problem go away? I mentioned this hidden in a thread a while ago, and am 
trying to get a reasonable fix.

Con

Re: 2.5.74-mm1 and Con Kolivas' CPU scheduler work

[Tom Sightler]

On Thu, 2003-07-03 at 19:49, Con Kolivas wrote:
> On Fri, 4 Jul 2003 02:36, ismail (cartman) donmez wrote:
> > Hi,
> >
> > With Con Kolivas' cpu patch I got many slow downs. When I try to run
> > CodeWeaver's CrossOver office box gets unresponsive at least for a one
> > minute. While starting KDE ( when splash screen is running ) mouse gets too
> > sluggish.
> 
> While your wm starting if your mouse is sluggish it is probably not a big 
> problem, and it is part of something I'm already working on. The codeweavers 
> problem is impressive though; I don't have codeweavers and have not seen 
> anything like it. To get a full picture it would be helpful if you could run 
> top as root say reniced to -13 just so it doesn't miss anything, and then 
> start codeweavers, watch top and tell me what is spinning away hogging the 
> cpu.

My guess is that he's seeing the same issue that I reported a few weeks
ago with Crossover plugin.  That report generated a fairly long thread
and a lot of patches and testing, but little actual progress.  Basically
wine seems to have multiple threads and a client/server architecture and
seems constantly spin on a pipe passing info back and forth to each
other.  It seems one thread gets all the attention and starves the
others although the others are critical for the main process to actually
make progress.

I've been wanting to do some testing with your patches so I'll give them
a spin tonight and see how this acts.

Later,
Tom

Re: 2.5.74-mm1 and Con Kolivas' CPU scheduler work

[Con Kolivas]

On Fri, 4 Jul 2003 02:36, ismail (cartman) donmez wrote:
> Hi,
>
> With Con Kolivas' cpu patch I got many slow downs. When I try to run
> CodeWeaver's CrossOver office box gets unresponsive at least for a one
> minute. While starting KDE ( when splash screen is running ) mouse gets too
> sluggish.

While your wm starting if your mouse is sluggish it is probably not a big 
problem, and it is part of something I'm already working on. The codeweavers 
problem is impressive though; I don't have codeweavers and have not seen 
anything like it. To get a full picture it would be helpful if you could run 
top as root say reniced to -13 just so it doesn't miss anything, and then 
start codeweavers, watch top and tell me what is spinning away hogging the 
cpu.

Con

> I tried Con's patch with 2.5.73-mm3 too and had same overall bad affects.
> 2.5.73-mm3 works very good.

Work with me and I'll try to improve your experience.

> Anyone has an idea whats might be wrong?

Con

2.5.74-mm1 and Con Kolivas' CPU scheduler work

[ismail (cartman) donmez]

Hi,

With Con Kolivas' cpu patch I got many slow downs. When I try to run 
CodeWeaver's CrossOver office box gets unresponsive at least for a one 
minute. While starting KDE ( when splash screen is running ) mouse gets too 
sluggish. 

I tried Con's patch with 2.5.73-mm3 too and had same overall bad affects. 
2.5.73-mm3 works very good.

Anyone has an idea whats might be wrong?

Regards,
/ismail

-- 
An Eye for an Eye will make the whole world blind - Gandhi