Hi,
we are currently hosting gameservers. We have 3 Dual Athlon MP servers.

We are having problems with the last one. By typing uptime, the system reports a load of 0.01 but when we look at the cpu usage, it should be higher. Something like 0.50 would be normal.

Also, by looking at cpu usage, it looks like the load is not well balanced between the cpus. CPU usage on our other servers look like this:

CPU0 50%
CPU1 43%

Which seems ok. But on the problematic server, CPU usage is more like:

CPU0: 73%
CPU1: 3%

Sometimes it's the other way around, CPU1 is more used but they never seem to be well balanced.

Server is stable and never crashes but we notice that the server sometimes hang up for a few seconds. While playing on a game server on that machine, from time to time, the game would freeze for 15 seconds.

I tried everything that came to my mind and I can't find the damn problem.

Any help would be much apreciated.

Thank you.

now game servers are out of my scope, but the issue of a cpu in a dual ( or quad ) setup being used exclusively, would seem to my a script that is not letting go and releasing the cpu back to the pool.

try this, next time run top against the cpu in question and see which process is hogging the cpu, try this for a week or 2, then you should have a clear idea of which process it might be. then you can go back to the game forums and ask about those processes.

Mike

I already did. The processes are a few "hlds" with is the game server. Hlds is not multi-threaded but since we run a few instance, they should be split between the two cpus and not hogging only one cpu. It should do the same as our other Dual AMD boxes.

I'm starting to think it is hardware related but which part could it be ?

/var/log/messages doesn't say anything abnormal.

I happen to be windows mostly, so to think unix resoultion is not so easy for me. The only things I can think of that will hog a CPU within a dual (quad ) envionment is

a) ram disk i/o
b) ide drive control card
c) bad network card
AHHH d) recheck the ram and make sure that's ok
e) try this if the above fails then look for updated drivers for each card you have and a bios upgrade

Thank you for your help Mike, it's much appreciated.

I didn't think a network card could hug the CPU. By seraching on the net, it looks like the network card we are using (3com 3c59x) is having some problems when used on a SMP system.

Some people suggested to upgrade the driver but 3com doesn't have the linux drivers available for download on their site for this model.

If somebody knows where I could get the latest driver for the 3c59x family ethernet card, please tell me ;)