On Tue, 4 Feb 2003, Kris Kennaway wrote: > On Tue, Feb 04, 2003 at 09:54:23PM -0800, Steve Kargl wrote: > > On Tue, Feb 04, 2003 at 09:38:18PM -0800, Kris Kennaway wrote: > > > I just booted a kernel with SCHED_ULE. It looks like there's a pretty > > > serious bug: > > > > > > PID USERNAME PRI NICE SIZE RES STATE TIME WCPU CPU COMMAND > > > 573 dnetc 139 20 1000K 804K RUN 1:29 85.94% 85.94% dnetc > > > 661 kris 96 0 2252K 1496K RUN 0:00 6.25% 6.25% top > > > 590 root 96 0 28620K 28128K select 0:04 3.12% 3.12% XFree86 > > > 641 root 120 0 4856K 4744K RUN 0:03 3.12% 3.12% make > > > > > > The make you see there was a 'make -j4' in /usr/src/secure. It has > > > been sitting there for about 5 minutes having done nothing other than: > > > > > > > Kris, > > > > How older is your src/ directory? I reported a similar > > problem to Jeff right after he committed ULE. He > > "fixed" the problem a couple days laters. I put "fixed" > > in quotes because after his fixes the system experienced > > 2 stalls under heavy load, but I couldn't prove it was > > ULE related (David Xu's KSE commit may have been involved > > in the stalls). > > I cvsupped and built kernel tonight. >
I may have broken the nice stuff when I was fixing interactivity. Or maybe I broke nice and interactivity when I was fixing SMP case. ;-) Anyway, I have a few ideas. \I'm going to play some more with the SCHED_STRICT_RESCHED stuff and make it automatic. I have some scripts that prove the nice computations as well. I'll rerun those. Cheers, Jeff To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
