[SunRescue] System Loads

[Dave McGuire]

On March 6, Paul Khoury wrote:
> Would anyone be able to tell me what system loads mean, like the following
> 
> # rup
>       cyprus    up 10 days, 14:03,    load average: 1.00, 1.00, 1.01
>       greece    up 15 days, 13:04,    load average: 1.06, 1.02, 1.02
>      lebanon    up  6 days, 12:06,    load average: 1.00, 1.00, 1.01
>       sparc2    up 66 days,  2:38,    load average: 1.09, 1.09, 1.10
>    macedonia    up 22 days, 12:49,    load average: 1.11, 1.09, 1.08
>       canada    up 15 days, 21:33,    load average: 1.16, 1.10, 1.09
>       jordan    up  2 days, 22:26,    load average: 1.28, 1.12, 1.09
> #

  The three numbers are the number of processes in the "run queue" (in
a state where they are eligible for cpu cycles and not waiting on
device I/O, etc) averaged over the last one, five, and fifteen minutes.

  As a system gets more and more work to do, more of these runnable
processes will get "backed up" waiting for cycles, more or less.

> What is a good and a bad load?

  It's not as clear-cut as that.  If you have one machine doing one
thing that's eating most of the cpu (running an rc5 client for
example) the load on the system will run around 1...on pretty much
any system.  It's kinda subjective, really...A load average of 8 or
10 may drag most PCs to their knees, where a well-stocked Sun4/6x0
with good high-bandwidth I/O controllers (IPI) and lots of memory
will usually chug along with a load of 30-40 without having any
trouble.


               -Dave McGuire