basic load average question... 3

[InigoMontoya]

The current load average on our sunfire v880 is as follows:

0.45, 0.30, 0.24,

Im not sure if this means it's running ok or if it's being over burdened. I'm not familiar on how to read these numbers. I got these figures from the 'top' utility. Can someone shed some light on this for me? Thanks.

 

[jimbopalmer]

Sure thing

this is your load average for 1 minute 5 minutes and 15 minutes
in the last 1 minute you were using 0.45 of a CPU, so if this is a single CPU box it is under half busy, if it is an 8 CPU box it is under 1/16 busy.

They give you different time frames so you can tell momentary spikes of load from a sustained load. Your 15 minute average shows you only 'need' about 1/4 of one CPU of the speed you are using. (slower CPUs would raise the average, faster CPU's lower it)

If you had a huge load your load average might be larger than the number of CPUs, I had a 4 CPU sun that would show a load avarage of 37 just before it locked up totally, I bought faster CPUs and learned to manage the peak load better.

As an aside, sar -q looks at this same info slightly differently, it reports the % of time when you do not have enough CPUs (the second column) and the number of programs waiting for a CPU when at least one is waiting (the first column)


I tried to remain child-like, all I acheived was childish.

 

[marrow]

You could also run "vmstat 5" (at 5 secs intervals) to view real time figures for system. Check Procs r b w, if all at 0, no CPU or Memory Problems. id = current CPU % availability. See "man vmstat" for details. This command is the quickest way to check a system's performance.

 

[marrow]

vmstat is one of the quickest ways to check a system
"vmstat 5" gives I/O, CPU details every 5 seconds, checks all disks for high usage, paging activity - sr (scan rates) - if continuously > 200 indicates memory heavily used, increases as free memory drops to as low as 500KB, if low values indicate VM system is running well; procs (process arrival rate) r b w w = swapped processes - if >5 not good, high number of swaps means shortage of physical memory
r = run queue indicates jobs queuing - if > 5 this means insufficient CPU power in the system
b = processes blocked waiting for disk, high block rate indicates need for more memory or faster disks/controllers.
Virtual memory and Swap Space will normally go down and then recover, if it continues to fall and doesn’t recover you may have a memory leak on the system. If po regularly reports high numbers, then disk bandwidth is being eaten by paging (pi not useful, but po - I don’t know if this still important or accurate) but “If this value is consistently non-zero, then there is a memory shortage, and the system is being forced to move sections of some programs to disk”. If the number of "swap-outs" is also consistently non-zero, then your system is desperately short of memory, and is moving whole processes to disk, also free memory, uptime, swap and iostat, mpstat

The values above are a bit out of date but are simply a guideline on where to look.

 

[daFranze]

load means:
The load average numbers give the number of jobs in the run queue averaged over 1, 5 and 15 minutes.
I am not familiar with the load, too. I use top or prstat to determine if the machine is on heavy load.
Or use the sdtperfmeter to have a graph?


Regards
-- Franz
Sorry I'm not a native spaeker, I'm from Munich, Germany - "Home of the Whopper", oh no, "Home of the Oktoberfest" ;-)