[Lula] system load averages

[Peter Benjamin]

At 04:54 PM 4/8/2007, you wrote:
>Hi list,
>
>I thought some of you gurus could make sense of this for me...
>
>Both 'top' and 'uptime' produce numbers for system load averages at
>one, five and fifteen minutes.  I've read a bunch of stuff about
>system load averages but it seems vague and contradictory.  What
>numbers would actually indicate that the system cpu was overloaded?

Anything over 0.00 ...  lol

I consider 8 or 9 to be bad.

I try to keep it under 2 or 3, but do not worry about those values.

Values around 4 to 5 start having me worried, and I look into it.

Values above 6 find me doing something, like killing a run away
process, or blocking an IP of a web spider, or at least slowing 
it down with robots.txt.

Values at 7 to 8 I find myself doing a killall for web CGIs.
Let the web surfer press Refresh/Reload to get a full page.
It's some spider or surfer (click the same link 10 times 
in a row as their computer or local net is busy) abusing 
my machine.  I do worry about ecommerce that might be lost
due to this drastic measure, but it only happens once or
twice a year.  Why killall?  Eventually, some limit is 
reached, like number of processes, open files, RAM, etc,
and the system freezes, or crashes.  :-(

The true test for overload is latency of services.
That would be true for any hardware, CPU speed, etc.

For a web server that would be the time period from the 
incoming IP packet request was received to the last byte 
of the last outgoing IP packet for that request leaving
the NIC.  Of course, you can measure it locally, or
remotely.  Most do remote, but that takes into account
the ISP (backbone) networks between the two machines.

I'd like to know what others do.

Hope this helps.