Re: Even CPU distribution in short time slice

[Aivils S. <ai...@la...>]

On Thu, 13 Jan 2005, Peter Williams wrote:

> Aivils Stoss wrote:
> > 
> > On Wed, 12 Jan 2005, Peter Williams wrote:
> > 
> > 
> >>Aivils Stoss wrote:
> >>
> >>>We can force 2.6.XX use small timeslices by
> >>>nice +19 all processes, but anyway multiple current
> >>>games runs more choopy as under 2.4.XX
> >>>
> >>>I see effective_prio() in zaphod patch uses
> >>>complicate alghoritm, which i cannot decipher :)
> >>
> >>If you want tasks to have shorter time slices using Zaphod then all that 
> >>is necessary is to set them using the file 
> >>/proc/sys/cpu_sched/time_slice which should be present if the kernel is 
> >>built with CONFIG_SYSCTL selected.  To set time slices to 10 msecs as 
> >>root do:
> >>
> >>echo 10 > /proc/sys/cpu_sched/time_slice
> >>
> >>You can set the time slice value anywhere between 1 msec and 1000 msec.
> >>
> >>Peter
> >>PS In future versions of Zaphod this functionality will probably move to 
> >>/sys (i.e. sysfs) instead of sysctl.
> >>PPS There is a primitive GUI at 
> >><http://prdownloads.sourceforge.net/cpuse/gcpuctl_hydra-v1.6.tar.gz?download> 
> >>which can be used to control various Zaphod parameters which effect 
> >>performance.
> >>-- 
> >>Peter Williams                                   pwi...@bi...
> >>
> >>"Learning, n. The kind of ignorance distinguishing the studious."
> >>  -- Ambrose Bierce
> >>
> > 
> > Ok, it turn on gaming again ! Thank You.
> > Nvidia colsed source drivers seems dislike some in zaphod.
> > I am under 2.6.10-ruby - console related patch and
> > patch-2.6.10-spa_zaphod_FULL-v6.1
> > 
> > time_slice=5
> > X0 - X server
> > 
> > Aivils
> > 
> > Unable to handle kernel paging request at virtual address fffffe9c
> >  printing eip:
> > c0119ad0
> > *pde = 00001067
> > *pte = 00000000
> > Oops: 0000 [#1]
<skip>
> > 
> 
> OK.  How frequently does this occur?  Also I am unable to test this as I 
> don't have any nvidea devices so would you be able to test any fixes?
> 
> Peter
> -- 
I test lightly only. Seems idle box  never oopses.
Oops pop up under load, if any of users run game. CPU ocupant remain live.
We have multiple video heads. 
Experiment is unclean, because vanilla 2.6.10 is untested.
I will try another day.

Before i tried out change vanilla-kernel DEF_TIMESLICE .
My box became unstable if DEF_TIMESLICE < 5 and  HZ==1000.

Aivils