I have two machines running Mandrake 9.x (either 9.1 or 9.2, I can't remember). One of them is at home, the other is at work. Both PCs are nearly identical Compaq Deskpro boxes, Intel 333MHz, built-in NIC.

The box at home I have been most active on. It is really stable, except under heavy network traffic (lots of shoutcast listeners, FTPing large files, etc) it simply drops off of the network. I don't see any really telling log messages, it just seems to forget it's ip (ifconfig shows no ip for eth0) and I can not get it to come back up.

I've tried using /etc/init.d/network restart, and that just times out at the point where it tries to get the ip. The only way I can get it back online is a reboot.

Well, my work MDK box has been rock solid, online for nearly a year straight with no problems. Yesterday I set up a test instance of Cruise Control on it, which is a J2EE build monitor, and set it to run overnight. This means it was checking out a large project from CVS every minute or so, and then attempting to compile it. This would be pretty heavy network traffic, as the project is large and takes nearly a minute to transfer.

When I came in this morning, the box was offline, exactly like what happens at home. /etc/init.d/network restart times out, unable to get back online. I needed to get the box back online ASAP, so I didn't have time to investigate, so I rebooted. After that everything seemed great.

Is there a known problem with MDK dropping offline under heavy network traffic? I've always assumed my PC at home had some problem with the hardware or something. But I cant believe both PCs would show the same symptoms, even though they are very similar, hardware-wise.

More than likely it's the built-in NIC. Do you know what chipset it has, or what kernel module loads for it?

I don't know what the actual network card is, but it looks like it is loading tlan... does that make sense?

Obviously, I'm not a Linux expert. I know enough to get frustrated.

TLAN is correct, that's a Texas Instruments ThunderLAN chipset that exists in many Compaq boxes.

See this post from 1999(!) that seems to describe a very similar problem:
http://www.uwsg.iu.edu/hypermail/lin...12.1/0020.html

Another interesting hit from Goolge is
linux24.sourceforge.net, under the "fixed" heading it lists:
Tlan and Epic100 crash under load
The last update of that page was in 2000, though.

So it does look like this was a known kernel/driver issue, but should have been fixed several years ago. MDK 9.x is more recent than that, so one would like to think it has the fix. Perhaps something specifically that Mandrakesoft did re-broke it?

Wow, thats very informative. Thank you.

I am actually pretty disapointed in Mandrake about this. Especially after talking my boss up about how easy MDK is to configure and run, and now I have to explain that I can't run our build monitor because it kills the nic driver.

Thanks again.