Server Load Question

daveirl

This post has been deleted.

Gerry

can you post up the output of the first say 10 lines from the "top" command, so we can see which process is causing all that load? It looks like some process has really gone crazy on your machine, a typical high load average for a 1 cpu box would be 3 or 4.

daveirl

This post has been deleted.

leeroybrown

load average: 5.60, 6.05, 6.44

Those load averages (above snippet from top) are on the high side but nothing unusual for a busy server.

That graph (I'm assuming MRTG) has obviously been scaled by a factor of around 100. The highest load average I've seen was around 60 on a dual CPU server with a few run-away processes and it only got that high because it was an IDE machine that dealt poorly with all the concurrent disk i/o generated.

92 processes: 86 sleeping, 6 running, 0 zombie, 0 stopped

A load average of 800 (as shown on that incorrect graph) would mean that at any given scheduling time slice there were 800 processes queued requesting processing at that instant. Now that would be scary...

daveirl

This post has been deleted.

bminish

daveirl wrote:

This post has been deleted.

It looks like there is a lot of swapping to disk going on, under those circumstances things can get pretty unresponsive since much of your 'memory' is actually hard disk.

I have 2 CC boxes and had a hand in setting up a third, I have never tried it with less than 256 Mb of ram and it certainly looks like ram is the bottleneck in your setup. Ram is cheap (even free) for older boxes but if you can't add more ram at least make sure that your hard drive(s) are running as fast as possible
hdparm

.Brendan

leeroybrown

load average: 5.60, 6.05, 6.44

Load average is a measure of the average number of processes queued in the scheduler for processor time slices over a period of time. The three load averages normally correspond to 1 minute, 5 minutes and 15 minutes. For example, your 1 minute load average of 5.60 means that the average number of processes waiting for CPU access at any instant in time during the last 60 seconds is 5.6. The 5 minute and 15 minute averages extend the premise and allow you to see the trend over the short term.

A load average of 1.0 would signify ideal CPU usage with full utilisation and zero waiting. When the load starts to get higher the time processes dealing with user interaction have to wait will increase making the server unresponsive.

Also, as Bminish mentions you're swapping a lot which is seriously affecting your performance and heavy disk access on IDE systems really hurts performance and creates an artificially high load average due to the delay it causes.

daveirl

This post has been deleted.

bminish

daveirl wrote:

This post has been deleted.

You could always offload the spamassin part of the job to another machine as it seems that spamassin is the main memory hog?
I have no experience of spamassin, I use Popfile which can run on another machine on the network

So you have a bit of experience with CC? I have to give it a big thumbs up in general. I tried to set this up with Mandrake and things before but CC made things so much easier without X and just generally being lighter.

I have been evangelising about CC for a while now on boards another much lighter firewall only product is M0n0wall which is the dogs bollux if you just want a standalone firewall / Bandwidth manager. M0n0wall is a particularly nice solution on Wrap boards with an old 8Mb CF card as the 'hard disk'

.Brendan