Subject: All meta-buffers in use, high load, no cpu activity
To: None <tech-kern@netbsd.org>
From: Wouter Schoot <wouter@schoot.org>
List: tech-kern
Date: 03/14/2004 10:20:20
Hi all,

Recently, I installed a new box (AMD 700, 384MB RAM, 200G Western Digital
disc, NetBSD 1.6.2). After reboot, it runs for a moment, but when I compile
something (take KDE3 for example - any big package seems to trigger this
behaviour), it quickly fills up all metadata buffers:

There are  4935 metadata buffers using            19740 kBytes of memory.
There are 11984 pages for cached file data using  47936 kBytes of memory.
There are  3653 pages for executables using       14612 kBytes of memory. 

File System          Bufs used   %   kB in use   %  Bufsize kB   %  Util %
/                         2467  49       19694  99       19712  99      99
/var                         3   0          17   0          20   0      85

Total:                    2470  50       19711  99       19732  99      99 

Top header gives:

load averages:  2.26,  1.54,  0.74
128 processes: 126 sleeping, 1 zombie, 1 on processor
CPU states:  1.5% user,  0.0% nice,  1.5% system,  0.5% interrupt, 96.5% idle
Memory: 107M Act, 1012K Wired, 15M Exec, 46M File, 229M Free
Swap: 701M Total, 701M Free 

So there's almost no activity on the cpu, while the load is high, and
discactivity seems low. I probably hit some buffer-limit, but I don't know
howto raise it. I noticed that the "4935 metadata buffers" always have the
same ammount, and never rises, changes.

Well, what happens may be obvious, compiling takes *ages*.
While KDE3 compiles, make extract in /usr/pkgsrc/mail/mutt takes over 30
minutes, however, the cpu time is only a few seconds.

Dmesg shows nothing unusual, nor does /var/log/messages, I use GENERIC kernel
that came with 1.6.2, it has 384 MB RAM. 

I had a look on /sys/arch/i386/include/param.h, which seems to define some
numbers, but I'm not sure if this is related.

I did read in the 1.6.2 release notes, that something had been changed over
the release, regarding to these buffers.

So is there anything that I can do to increase the number of those buffers,
or may the problem be related to something else ?

Thanks in advance,

Wouter

-- 
Wouter Schoot (wouter@schoot.org)
Website: http://www.schoot.org
UIN# 42109851