Re: ULE/SCHED_SMP diff for 7.0

On Thu, Jul 19, 2007 at 08:00:34PM +0200, Rene Ladan wrote:
> Jeff Roberson schreef:
> > On Thu, 19 Jul 2007, Rene Ladan wrote:
> > 
> >> Jeff Roberson schreef:
> >>>
> >>> On Wed, 18 Jul 2007, Rene Ladan wrote:
> >>>
> >>>> Jeff Roberson schreef:
> >>>>>
> >>>>> On Wed, 18 Jul 2007, Rene Ladan wrote:
> >>>>>
> >>>>>> Jeff Roberson schreef:
> >>>>>>> http://people.freebsd.org/~jeff/ule.diff
> >>>>>>>
> >>>>>>> This patch is scheduled for inclusion in 7.0.  I would like
> >>>>>>> anyone who
> >>>>>>> cares to run it to validate that it does not create any stability or
> >>>>>>> performance regression over the existing ULE.  This patch
> >>>>>>> replaces ULE
> >>>>>>> with SCHED_SMP, which will now no longer exist as a seperate fork of
> >>>>>>> ULE.
> >>>>>> [..]
> >>>>>>
> >>
> > 
> > Thank you.
> > 
> > This is definitely a KSE/ULE problem.  Can you try
> > http://people.freebsd.org/~jeff/sched_ule.c?  I have made some changes
> > that may fix this.
> 
> cvsuped today at 17:23 UTC and replaced sched_ule.c afterwards.  The almost-instant
> panics are gone.  Also, the oddity I reported about ULE/BOINC on this list earlier
> has been fixed by the new sched_ule.c, I've noticed however that the WCPU field in
> top sometimes exceeds 100% (this wasn't the case with sched_4bsd).
> 
> Thanks,
> Rene
> -- 
> GPG fingerprint = E738 5471 D185 7013 0EE0  4FC8 3C1D 6F83 12E1 84F6 (subkeys.pgp.net)
> 
> "It won't fit on the line."
> 		-- me, 2001
>

Patch works:

This patch settled all issues with ULE on my dual amd64/2800. No more sudden
jumps in 'load' to 50 or higher. (at worst it hit something absurd like 
7000) No more crashes (knock on wood) that were apparently associated with 
this. Thanks for the good work.

Looking out for:

There will be one area I will be looking out for: Some applications like 
'Evince' don't allays kill themselves off completely when finished, leading
to some of the excessive load. Another case is when Linux-mozilla crashes 
from running 'Flash' routines. The messages log says it exited, ps tells 
another story. It is not yet known if this is fixed. Hope so.

Linux issues:

There was an earlier post about Java on Linux being unstable. I'm using the 
new fc6 libs. This is good and i'm able compile almost everything from source
for BSD and my more 'down to earth' Linux desktop. I've built the full set 
of libs (except glibc) and they work fine too. (Used a minimal install from
Gentoo.) There are, as expected, a few glibc issues. If make(1) wants to 
delete a file that is at the end of a long path, it will fail. Same for 
chmod and other utils that must 'reach' for a certain file. (to quote from
the error message) It is possible to make this all work by placing a static
util build (like rescue) in the Linux. I don't really want to do that. 
Bottom line: It would be nice to have a jdk again....   

Thanks again,

Bill Squire