From: Joachim Schipper (j.schippermath.uu.nl)
Date: Fri Dec 02 2005 - 04:15:27 CST

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On Thu, Dec 01, 2005 at 09:58:00PM -0200, Beto wrote:
> Hi friends,
> I got a strange high cpu usage on my server with openbsd 3.8 release.
> I have on my lab 20 machines using ldap to authenticate and squid proxy
> transparent to access the internet. Their homes were mounted by nfs, however
> my system was crashing frequently and now they just login and don't mount
> anything. It's a hub and I have 2 logical networks.
> Althought i have just 20 machines there are 3000 users, (but they don't
> login at the same time).
> My server is an old machine but i thought it is enough for it.
> It is an Intel pentium 3 833Mhz 512Mb ram
> My server runs:
> dhcpd
> ntpd
> named
> squid
> squidGuard
> slapd
> nfsd
> sshd

While I don't know or run all this software, it should be okay. OpenLDAP
is rather heavy, but if used only for authentication this should not be
a problem.

> I runned an tcpdump and the trafic is normal.
> Some one have any idea? I realy need some help and I did not found anything
> about it.
> Thank you.
>
> What my top show:
>
> load averages: 2.00, 1.82, 1.69
> 21:02:12
> 40 processes: 3 running, 36 idle, 1 on processor
> CPU states: 53.8% user, 0.0% nice, 23.9% system, 13.1% interrupt, 9.2%idle
> Memory: Real: 65M/115M act/tot Free: 383M Swap: 0K/700M used/tot
>
>
> load averages: 1.50, 1.60, 1.60
> 21:33:01
> 40 processes: 3 running, 36 idle, 1 on processor
> CPU states: 59.3% user, 0.0% nice, 25.3% system, 9.3% interrupt, 6.1%idle
> Memory: Real: 65M/115M act/tot Free: 383M Swap: 0K/700M used/tot
>
> PID USERNAME PRI NICE SIZE RES STATE WAIT TIME
> CPU COMMAND
> 14610 named 64 0 3032K 3384K run
> - 68:39 50.34% named
> 26369 root 58 0 78M 14M run
> - 46:56 30.81% slapd
> 7008 _syslogd 2 0 352K 604K run
> - 28:11 2.10% syslogd
> 24914 _squid 2 0 35M 37M sleep
> poll 3:13 0.68% squid
> 7849 root 2 0 332K 76K sleep
> nfsd 0:33 0.00% nfsd
> 25407 _squid 2 0 580K 840K sleep
> netio 0:13 0.00% squidGuard
> 30316 root 2 0 308K 508K idle
> netio 0:05 0.00% syslogd
>
> I had cutted the rest of what it printed.
>
>
> My dmesg:
>
> PACE to TIME
> /var: optimization changed from TIME to SPACE
> nfsd send error 55
> /var: optimization changed from SPACE to TIME
> /var: optimization changed from TIME to SPACE
> /var: optimization changed from SPACE to TIME
> nfsd send error 55
> nfsd send error 55
> nfsd send error 55
> nfsd send error 55
> nfsd send error 55
<more along this vein, and a couple of boots that look ok to me>

Looks like /var is very full, better get some more disk.

Also, nfsd is complaining - Google yielded that error 55 may well be
from /usr/src/sys/sys/errno.h - i.e., ENOBUFS /* no buffer space
available */.

I wouldn't know *why* there is no buffer space available, as I don't use
NFS - it might well be in TFM or TFF, and if not, maybe someone here
knows. Just a guess, but NFS may use something like shared memory, which
can be exhausted much more easily than physical memory. There are some
sysctl knobs to fix this.

                        Joachim

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]