*BSD News Article 75082


Return to BSD News archive

Path: euryale.cc.adfa.oz.au!newshost.carno.net.au!harbinger.cc.monash.edu.au!munnari.OZ.AU!news.ecn.uoknor.edu!qns3.qns.net!imci4!newsfeed.internetmci.com!uuneo.neosoft.com!news.sesqui.net!uhura.phoenix.net!usenet
From: leigh@phoenix.net (Lucas Wolenczak)
Newsgroups: comp.unix.bsd.freebsd.misc
Subject: ircd leak in FreeBSD
Date: Tue, 30 Jul 1996 19:21:57 GMT
Organization: Phoenix Data Net (713) 486-8337 http://www.phoenix.net
Lines: 29
Message-ID: <4tln4t$5e7@uhura.phoenix.net>
NNTP-Posting-Host: dial29.phoenix.net
X-Newsreader: Forte Free Agent 1.0.82

I am running the FreeBSD 2.1.0-RELEASE version - the machine houses
The DALnet IRC Services code which takes up approx 41M - the machine
however is 96 M.  The ircd is having problems with allocating memory.
It grabs what memory it needs and when it's done with it, it seems to
forget to give it back, thus ircd which shouldn't go any higher than
12M MAX - ends up around 40M.  It keeps climbing so that if the
machine isn't restarted every couple of days - it runs out of memory
and swap and the machine eventually dies.  Only machines on DALnet
experiencing this prob thus far are FreeBSD machines.  there is on
FreeBSD machine running 2.1.5 but it's a new server and unknown yet if
it has been experiencing this mem leak as well.  We've tried shutting
down the ircd code and restarting it to reset the memory but even then
it doesn't seem to let go - only a full reboot resets the memory for
use again.  Would upgrading to 2.1.5 fix this problem? Is there
anything in ircd or FreeBSD to slow down or teminate this mem leak
problem?  We've gone into /sys/sys/types.h and reset FD_SETSIZE=1024
to allow for 1024 in ircd and set the makefile to compile with
FD_SETSIZE=1024 - we've verified that has worked as this machine under
64M before ran beautifully with 1015 users (with services it cut out
there till we got more memory).  Something that I have noticed is that
the more users on the server - the faster the ircd goes out of control
with memory hogging.  This machine is one of dalnet's major hubs -
it's rather debilitating to only be able to withstand 80 clients 8-13
servers while running services and still having to reboot every 2-3
days to keep the machine from dying.  Any help on this issue would be
greatly appreciated.  

-Leigh