Date: Wed, 11 Mar 1998 13:48:39 -0500 (EST) From: Peter Dufault <dufault@hda.com> To: mi@aldan.algebra.com (Mikhail Teterin) Cc: freebsd-current@FreeBSD.ORG Subject: Re: idprio/rtprio Message-ID: <199803111848.NAA26056@hda.hda.com> In-Reply-To: <199803111740.MAA00835@rtfm.ziplink.net> from Mikhail Teterin at "Mar 11, 98 12:40:58 pm"
next in thread | previous in thread | raw e-mail | index | archive | help
> Is the <subj> broken right now? > > I upgraded to current as of Mar 8, but only rebooted to the new > kernel today. > > First, contrary to what the manpage says, an ordinary user can no > longer run idprio. > > Should I file a PR? The code comment shows that is done on purpose to avoid priority inversion. > Now, something less obvious. I run rc5des (start it as myself, then > make it idprio as SU). The machine slows down more then it used to > under the same conditions. Top reports the rc5des' nice level as 31: > > CPU states: 1.1% user, 97.4% nice, 1.1% system, 0.4% interrupt, 0.0% idle > Mem: 3444K Active, 38M Inact, 14M Wired, 5112K Cache, 7620K Buf, 1656K Free > Swap: 96M Total, 3320K Used, 93M Free, 3% Inuse > > PID USERNAME PRI NICE SIZE RES STATE TIME WCPU CPU COMMAND > 729 mi 101 31 1060K 292K RUN 14:53 97.35% 97.35% rc5des > 424 mi 2 0 4216K 3392K select 0:54 0.38% 0.38% Xaccel > 824 root 29 0 820K 760K RUN 0:00 0.84% 0.31% top > 455 mi 2 0 264K 496K select 0:00 0.04% 0.04% xload > > Load is 1.44-1.55 -- it was at the strict 1.0 before, unless I was > running something else that was cpu. hungry. > I broke it at 1.47 of kern_synch - idle priority processes are not getting preempted until the next round robin interval instead of as soon as a higher priority process is runnable. I did this locally a while ago so I have to dig up why I added that test to only do need_resched for the same type of process. Eliminating that check in maybe_resched for the scheduler types should set things back the way they were. I'll run a few tests with that off and see why I added it. Peter -- Peter Dufault (dufault@hda.com) Realtime development, Machine control, HD Associates, Inc. Safety critical systems, Agency approval To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199803111848.NAA26056>