Router dies after while and is not reachable

McBrain · 10-29-2003, 05:15 PM

Hi there,

I am using a Slackware 9.1 system on a very old pentium PC as a router and gateway. The connection to the net is done over a fritz dsl card and my normal desktop PC is connected to the router via an ethernet card.

It happens regular, that after some time (period changes) the router seems to die. It is not reachable anymore, but also logfiles are not written anymore. So it seems it really dies.
But this seems only to happen when my eDonkey is running on the desktop system (independent from the operating system, tried Windows and Linux). So it seems that something in this protcoll kills it! But I thought I implemented a firewall (mainly denying everythink incomming on the ppp0 device expect the eDonkey ports which are forwarded to the desktop).

Any idea what the problem is or how I can come closer to the solution?

Thanks and regards!

zaphodiv · 10-31-2003, 10:46 AM

what does the router machine show on the screen?
Does it still respond to the keyboard?

clacour · 11-01-2003, 01:09 AM

First thought is that the hardware is old and tired and any kind of serious load kills it.

Somewhat related idea: bug in the driver for one of the two network devices is killing the machine, and increased activity triggers the bug.

A few ideas for tracking down the problem:

Write a small script to simply log the fact that the box is up. Pay careful attention to the time (and make sure ahead of time that both boxes have good time) when you reboot it. If the script continued to log right up until the time you rebooted it, the box isn't completely dead.

Quote:

#!/bin/bash

while :
do
date >> /tmp/lastup
sleep 1
done

Also look for gaps in that. On occasional one of one second is nothing to worry about, but if there are long gaps, or lots and lots of one and two second gaps, that means the systems bogged down very heavily.

If it stops when you first see the symptoms of the box's "death", then it really is locking up hard. That points to either a serious kernel bug (almost certainly in one of the network drivers) or hardware. (My money would be on hardware.)

If the box dies slowly, a trick I've found to be extremely useful is to set up a bunch of commands (examples: "vmstat 5 5", "netstat", "ps -ef", "tail -20 /var/log/messages","ifconfig", etc.) that you would LIKE to had the chance to run when the thing started screwing up. Run this script every 5, 10, 15 minutes -- whatever's appropriate to how long it takes to get messed up. (The closer together you're going to run it, the shorter it should be.)

Finally, put an entry in /etc/syslogd.conf that says "*.debug" and point it to some file (probably would be better not to make it your /var/log/messages file). That might get you some info on the problem that you currently can't see.

Hope this helps,

CHL