*BSD News Article 8826


Return to BSD News archive

Path: sserve!manuel.anu.edu.au!munnari.oz.au!ariel.ucs.unimelb.EDU.AU!werple.apana.org.au!news
From: andrew@werple.apana.org.au (Andrew Herbert)
Newsgroups: comp.unix.bsd
Subject: Re: Occasional system hangs
Date: 14 Dec 1992 12:52:54 +1100
Organization: werple public-access unix, Melbourne
Lines: 44
Message-ID: <1ggpdnINN63l@werple.apana.org.au>
References: <andrewh.724059111@bruce.cs.monash.edu.au> <1gaclpINN84v@hrd769.brooks.af.mil> <1992Dec11.210430.17335@coe.montana.edu> <Bz495r.Byn@ns1.nodak.edu>
NNTP-Posting-Host: werple.apana.org.au

Mark Tinguely <tinguely@plains.NoDak.edu> wrote:

> Did all of you with 16+ Megs of RAM apply the patch to restrict the size of
> bufpages (Terry Lambert's patchkit patch #2.

Yup.  I'm also currently using 4MB of kernel memory to avoid the kmem_map
exhaustion problems I was still getting even after applying the patch.  Works
well, despite being somewhat wasteful!  I've applied all the relevant beta
patchkit fixes (thanks Terry!), plus most (reasonable looking :) fixes posted
to comp.unix.bsd.  Plus a few from bde@runx.oz.au to prevent hangs and
panics in the pty and 387 handlers respectively.  And a fix for an ICMP
redirect problem that caused 3 out of the 6 panics that have occurred since
I installed ddb and started tracking them down about a week ago (just about
to post that one...).

> Besides, there are small memory leaks in the kernel; Programs that fork a lot,
> cause the system to hang (make and apparently news); Big programs like X

FYI, INN doesn't fork much at all.  It is big though - here is what ps -alxw
has to say about it:

  UID   PID  PPID CPU PRI NI   VSZ  RSS WCHAN  STAT TT       TIME COMMAND
  200    75     1   0  28  0  4828 2928 -      Rs   ??   13:43.31 in.innd -p4

Are these small memory leaks you mention pages being lost in the VM code, or
kernel malloc()/free() related?  Since Julian's fix for the raw socket problem
that was hogging malloc(x, MBUF) space when ping or traceroute were run, I
haven't noticed any significant malloc leaks.

> I agree that a kernel fix is better than a symptom avoidance technique.
> The kernel change is massive. The VM allows over extention of the swap
> backstore (and this can lead to the system freeze -- all of the swap
....
> are put on swap backstore, but backstore is full ... system will work the
> drive a little but will eventually freeze.

Thanks for the explanation - it clarified a number of things for me.

So would you agree that adding enough RAM to avoid paging/swapping most of
the time would help avoid this problem?

cheers,
Andrew
andrew@werple.apana.org.au, andrewh@cs.monash.edu.au