What do the server load average numbers mean, specifically? Is that a percentage of the use of the processor? So a server load of 2.00 would mean that 2% of the processor's potential is being used?

1 is 100% for single processors
2 is 100% for dual processors

I see, thank you! :)

Is this true? I only ask because we had one customers P3 filled with run away cgi scripts and the load was 160 - yet the server still stood up -

thats 1600% of the CPU going by your rule....

I beleive those figures are off. My server which hosts a few of my personal sites, all are designed in html, each with a form-mail script. No databases, no CGI scripts, pages are all size optimized. Each site uses about 5-10GB per month. Yet my server load average is .40-.50.

The load average isn't the percentage of cpu usage, its the amount of processes that are currently running. I can't remember for sure how exactly it works at the moment, but try searching google. "Google, don't administer a server without it."

From the man page:

The load averages are the average number of process ready to run during the last 1, 5 and 15 minutes.

If I remember correctly, load average is the average number of porcess over a fixed period that tried to run.

For example, a load of 1 over the last 5 minutes means that there were 1 process running over the last 5 minutes. 2 would mean that 2 processes were running (presumably on a uniprocessor machine) means the two processes (not any two in particular) were fighting for CPU time. If you have less than 1 on a uniprocessor, that means it is (in a sense) underused.

Peter

(looks like RutRow and I were writing at the same time)

My understanding:

Load average is the # of processes waiting for resources - disk I/O, memory, or CPU.

So a load average less than 1.0 means that no processes are piling up in the run queue, whereas anything higher means that processes are having to wait, ready to run, for resources to become available.

Yet another different opinion here goes:

I believe it's the number of simultaneous requests the cpu is trying to process.

AcuNett, please define "requests". Do you mean Interrupt Requests?

The load average definitely is not just any kind of wait for any kind of resources; it is the wait for CPU. If a process is waiting for an IO request (net, disk, etc...), it cannot simoultenouly be waiting for the CPU. In fact, a process is in the run queue only if it is not waiting on any kind of resources. (Idle processes usually wait on a timer or an IO event).

Also, memory is not a "waitable" resource. You access memory when you get CPU time.

I am not linux kernel expert, but I did have some understanding of kernel design.

Peter

ckpeter,

Your comments make sense.

One interesting phenonmenon with regards to loadavg that I have noticed:

If you're compiling a bunch of packages and loadavg pops up to 3 or 4, the command line gets choppy and laggy.

But, I've seen loadavg go to 20+ due to NFS timeouts (Apache requests piling up, unable to contact the fileserver), and the command line is still responsive.