Date: Fri, 28 Jun 2002 11:25:48 -0700 (PDT) From: Matthew Dillon <dillon@apollo.backplane.com> To: Wm Brian McCane <root@mccons.net> Cc: <current@FreeBSD.ORG> Subject: Re: memory/swapping problem Message-ID: <200206281825.g5SIPmPm006258@apollo.backplane.com> References: <20020628080323.U91563-100000@fw.mccons.net>
next in thread | previous in thread | raw e-mail | index | archive | help
The amount of wired memory could be unrelated to the problem. Memory
associated with the buffer cache will be wired, but it is limited to
the size of the buffer cache.
(It's also possible that there is indeed some sort of wired memory leak).
p.s. I do not recommend using -current on a production system.
-Matt
Matthew Dillon
<dillon@backplane.com>
:On Fri, 28 Jun 2002, Anthony Jenkins wrote:
:
:>
:> Wm Brian McCane wrote:
:>
:> >I am having a problem with a new server that I am setting up to host my
:> >UseNet news and email.
:> >
:> >Machine:
:> > 1.26GHz 512MB/L1 Cache
:> > 768MB PC133 Memory
:> > 4x60GB UDMA100 IDE Drives
:> > 4x36GB 160MBit SCSI Drives
:> >
:> >OS: 5.0-CURRENT (yes I know, but I have been CURRENT since 386bsd 0.1)
:> >Software:
:> > INN 2.3.3
:> > Sendmail w/amavisd milter
:> > ipop3d
:> >
:> >This machine has been running for about 2 months on CURRENT with no
:> >problems, but I recently upgraded the CPU from an 866 to the new one. At
:> >that time, I also planned to replace the motherboard with a new Intel SAI2
:> >dual processor board, add a 2nd processor and drop in 3x512MB ECC
:> >Registered DIMMs. So I built and installed a more current CURRENT and
:> >added SMP to the config for my kernel. The motherboard was defective, so
:> >I put back in the original board and memory chips, while I wait for the
:> >replacement board. The 2nd night after the build was done, I started
:> >having memory/swap/wired problems. I rebuilt the kernel for UP operation,
:> >but the problems have not gone away.
:> >
:> >Machine averages about 95% idle all day except during news expiration.
:> >For the first day, it averages around 45M wired. 'inn' process is about
:> >137MB RSS. During expiration there is an understandable spike in CPU and
:> >Swap activity. 'expire' process is about 259MB RSS. When the expiration
:> >completes there is about 254MB still wired. The second night it goes to
:> >500+MB wired and the system never finishes expiration before I get a call
:> >that email is down. I have actually gone out and killed every process
:> >running on the machine that I can (including inetd, syslogd, cron, sshd,
:> >sendmail and inn). The wired memory is never released until I reboot the
:> >machine.
:> >
:> The only issue that I've had that remotely resembles that involes CPU
:> usage, not memory. Assuming your statistics are from top(1), can you
:> try 'top -S' to display what system process resources are also? I'm
:> betting it's some system task, since your killing user processes has had
:> no effect. (btw my issue is the "irq10:" task seemingly locked in
:> *Giant state and sucking >50% the CPU/WCPU measure.)
:>
:> --
:> Anthony Jenkins
:>
:Anthony,
:
: Tried this, nothing was showing huge memory consumption. I had to
:reboot the machine anyway, so while it was down I put in another 256MB of
:memory and manually ran the expire. No huge cpu/swap spike, and no
:increase in wired memory after the expire. Looks like the wired memory
:problem is tied to the swapping in some way, based on empirical data.
:
:- brian
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?200206281825.g5SIPmPm006258>