I have a problem with a server that is suddenly freezing, apparently
when running out of RAM. There is no message in the syslog or any
other log, the server just stops. I have monitored the resource use
with top. Last time the server froze the resource use was like this:

Mem: 515904K av, 504784 used, 11120K free, 4300K buff
Swap: 52207K av, 0K used, 522072K Free, 307196K cached
91 processes was running and the load was 0.02,0.01,0.00.

I have two theories about this; one is that swap isn't working
properly (because it doesn't use any swap when it hangs) or there is a
memory leak somewhere (?).

Any theories about this? I have changed RAM, video card and NIC
without any improvements. The server has been working ok about 7-8
months before this started to happen. The server hangs with both
kernel 2.4.19-33 and 2.4.18-8. (Mandrake Secure precompiled). Any tip
that could lead me in the right direction would be appreciated.

--
Sincerely,
David Karlsen

Something else is most likely causing the crash. Linux will attempt to keep
file references, etc, in RAM until it needs the RAM.  I believe its based on
FIFO.  So your numbers re: memory are not really applicable to your theory.
Try dmesg to see if anything is going on. Otherwise, look in your /var/log
directory for failures.

ken k

F.

--
Serwis Usenet w portalu Gazeta.pl -> http://www.gazeta.pl/usenet/

Can you double your ram and get rid of your swap?  We have 512M ram
and no swap on our server, it's been up 143 days with 80% max memory usage.

What exactly is this server used for?  What applications are running.
What services?  What are the syslog saying for the kernel etc, when
the systems starts to go south?  What's the status of your mounted
file systems?  Which ones are full?  Which are having the highest
usage stats?

Are you running the kernel with the overload parameters turn on?

Have you tried turning the swapoff and back on?   Created another swap
file, and tried using in?  Tried using the partition, and file at the
same time(better io or worse?) Changed the bdflush/buffermem/filemax
kernel parameters?  What do you have (ulimit) set to try ulimit -a and
find out?

Got Lsof?  Who's listening?  What are your netstats coming back with?
Do have have bandwidth & QOS turned on? Done network security limit
attacks by turning on/off kernel network options?

Have you tried a regular out of the box kernel?  Have you got hdparm
setting your drives to some odd parameter?  If you try setting umask
IRQ back to 0, and test (-acd) to make sure they are not interfering
with performance.
Find out if your load is realted to a user, or application.  See if
there is a relations some where.  User logs in and 2 hrs later the
system goes south.

Look into the Securing and Optimizing Redhat a good portion of that
will be helpful.

I would definitely try this one *FIRST*. In my experience things like
this are quite often caused by these kind of 'NO, that can't be, can
it?' whatchammacallits. To high a processor temp with 'freeze' you
machine in such a way that NOTHING (incl. keyboard, mouse) execpt the
power switch will respond.

R.A. Reissaus
Risdi, Amsterdam, the Netherlands
Ibm/Informix Consultants
Telefoon: +31(0)20 4942174      : Phone
Mobiel  : +31(0)6 51185352      : Mobile
Fax     : +31(0)20 4940795      : Fax  
E-mail  : r.a.reiss...@risdi.com
Website : www.risdi.com