torriem at gmail.com
Tue Sep 30 16:54:36 MDT 2008
Andrew Jorgensen wrote:
> Okay folks, I'm going out on a limb and admitting to some ignorance
> here. Suppose I have a high load average on a server, let's say 20,
> how do I tell what's really going on? I understand that load means
> that there are processes waiting for some resource but how do I see
> what resources they are waiting for? We don't want to go buy more RAM
> and then find out that we had plenty of RAM, for instance.
When the load is high, the first thing to check is the amount of swap
being used. If the swap is exceeding 50%, I've found that Linux tends
to thrash a lot. In this case adding RAM will help, although if this is
not a normal server condition, maybe you have a process with a memory leak.
If not, then check your disk i/o stats. If the disk is maxed out, or
even in action the majority of the time, then your load average is due
to disk wait, which is a hard one to solve. Sometimes a better i/o
scheduler can improve things. Sometimes faster disks.
If it's none of these things, then perhaps it's the network that
processes are waiting on. This isn't necessarily a bandwidth issue
either. If NFS is slow, the load average will shoot up, for example.
So load is caused by I/O problems, whether disk or network.
More information about the PLUG