From nobody@FreeBSD.ORG Thu Aug 19 03:01:22 1999 Return-Path: Received: by hub.freebsd.org (Postfix, from userid 32767) id 5A55715085; Thu, 19 Aug 1999 03:01:22 -0700 (PDT) Message-Id: <19990819100122.5A55715085@hub.freebsd.org> Date: Thu, 19 Aug 1999 03:01:22 -0700 (PDT) From: holger.kipp@solit-ag.de Sender: nobody@FreeBSD.ORG To: freebsd-gnats-submit@freebsd.org Subject: niced jobs don't behave really nice X-Send-Pr-Version: www-1.0 >Number: 13252 >Category: kern >Synopsis: niced jobs don't behave really nice >Confidential: no >Severity: non-critical >Priority: medium >Responsible: freebsd-bugs >State: closed >Quarter: >Keywords: >Date-Required: >Class: change-request >Submitter-Id: current-users >Arrival-Date: Thu Aug 19 03:10:00 PDT 1999 >Closed-Date: Sun Aug 5 02:46:40 PDT 2001 >Last-Modified: Sun Aug 5 08:40:02 PDT 2001 >Originator: Holger Kipp >Release: 2.2.8 >Organization: SolIT AG >Environment: FreeBSD intserv 2.2.8-RELEASE FreeBSD 2.2.8-RELEASE #0: Tue Jan 26 23:44:04 CET 1999 root@intserv:/usr/src/sys/compile/INTSERV i386 >Description: Niced jobs use up to much CPU time if other jobs are running (or are supposed to be running instead). >How-To-Repeat: Run some timecritical software (eg amanda) and install some other jobs that are supposed to _really_ run in the background (ie start them with nice -19). Specifically suited are CRACK and SETI@home. Then have a look at the jobs with top and see the niced jobs running with > 80% CPU usage even though amanda with dumper and gzip are running... >Fix: No idea, but under my Linux implementation "Linux clausthal 2.0.36 #5 Wed Jan 20 14:23:31 MET 1999 i686 unknown" I tested this with SETI@home with nice -19, and there was no slowdown whatsoever noticeable if running other processes. The niced SETI@home job is really only using the CPU-time the idle-process would otherwise use. >Release-Note: >Audit-Trail: State-Changed-From-To: open->closed State-Changed-By: sheldonh State-Changed-When: Thu Aug 19 04:45:05 PDT 1999 State-Changed-Why: Duplicate of PR 12381. This problem is fixed in RELENG_3, in rev 1.16.2.1 of src/sys/i386/isa/intr_machdep.c . State-Changed-From-To: closed->suspended State-Changed-By: sheldonh State-Changed-When: Thu Aug 19 07:21:19 PDT 1999 State-Changed-Why: As Bruce Evans points out, this problem still exists in RELENG_2_2. I'm leaving this suspended since RELENG_2_2 isn't completely dead. State-Changed-From-To: suspended->closed State-Changed-By: jon State-Changed-When: Sun Aug 5 02:46:40 PDT 2001 State-Changed-Why: RELENG_2_2 is completely dead, or so we hope. http://www.FreeBSD.org/cgi/query-pr.cgi?pr=13252 From: Bruce Evans To: Cc: , Subject: Re: kern/13252: niced jobs don't behave really nice Date: Sun, 5 Aug 2001 22:40:56 +1000 (EST) On Sun, 5 Aug 2001 jon@FreeBSD.ORG wrote: > Synopsis: niced jobs don't behave really nice > > State-Changed-From-To: suspended->closed > State-Changed-By: jon > State-Changed-When: Sun Aug 5 02:46:40 PDT 2001 > State-Changed-Why: > RELENG_2_2 is completely dead, or so we hope. This is the same bug as in the suspended PR 12381. The fix referred to in the followup (rev.1.16.2.1 of isa/intr_machdep.c) is for a different bug. See PR 12381 for a much longer followup including a correct reference to the fix. Current status of this bug: - never fixed in any 2.* or 3.* release. - PR 12381 hidden^Wsuspended awaiting commits to RELENG_3. - fixed in all 4.* releases by inheriting rev.1.83 of kern_synch.c (etc.). - re-broken in -current in rev.1.90 of kern_sync.c (etc.). - made worse in -current by SMPng prioritization changes (these squeeze the priority space, and the priority space was already too congested to work properly). Bruce From: Jonathan Chen To: Bruce Evans Cc: holger.kipp@solit-ag.de, freebsd-gnats-submit@FreeBSD.ORG Subject: Re: kern/13252: niced jobs don't behave really nice Date: Sun, 5 Aug 2001 11:34:50 -0400 On Sun, Aug 05, 2001 at 10:40:56PM +1000, Bruce Evans wrote: > This is the same bug as in the suspended PR 12381. The fix referred > to in the followup (rev.1.16.2.1 of isa/intr_machdep.c) is for a > different bug. See PR 12381 for a much longer followup including a > correct reference to the fix. > > Current status of this bug: > - never fixed in any 2.* or 3.* release. > - PR 12381 hidden^Wsuspended awaiting commits to RELENG_3. > - fixed in all 4.* releases by inheriting rev.1.83 of kern_synch.c (etc.). > - re-broken in -current in rev.1.90 of kern_sync.c (etc.). > - made worse in -current by SMPng prioritization changes (these squeeze > the priority space, and the priority space was already too congested > to work properly). Oops, this will teach me to never blindly trust what somebody else wrote a long time ago... In any case, we can probably leave this closed as it is a duplicate of another PR. Thanks for catching this, Bruce. -Jon >Unformatted: