[Asterisk-Users] CPU load (was: dimensioning: Where is the CPU vsAsterisk load table)

[Mimmus]

> -----Original Message-----
> From: asterisk-users-bounces at lists.digium.com 
> [mailto:asterisk-users-bounces at lists.digium.com] On Behalf Of 
> Simone Cittadini
> Sent: Friday, January 13, 2006 12:00 PM
>
> Erick Perez ha scritto:
> 
> >
> >-And the most important I read was: Keep load under 5 in single CPUs 
> >and 10 in dual CPUs (didn't mention dual cores in the article).
> >
> >
> >  
> >
> That seemed to me a lot, so i googled around a little trying 
> to understand the true meaning of those numbers :
> I'll sum up here what I've found, sparing you the formulae 
> (look for "linux load average" "neil gunther") First of all 
> the sampling of cpu load gives more weight to recent samples, 
> so is better to look at the third value, average in the last 
> 15 minutes, without being scared by high "punctual" values. 
> Following "what the gurus says" the value should be kept 
> below 3, or below the number of cpus, given what we are 
> measuring (the number of process ready and waiting to be 
> executed), those values means to me "a rule of thumb" and 
> "make no one wait to do his job". It's not a lot of meaning, is it ?
> What I suppose we want to say is "when I start hearing the 
> calls bad ?", like gamers don't care about FPS but want to 
> know "which graphic card I have to buy to frag aliens 
> smoothly ?". I'm not a C programmer so I don't know asterisk 
> internals, what I'll say now maybe is totally nonsense, I 
> leave the sensate replies to the community.
> If I have an asterisk process waiting, is sensate to state 
> that if it waits too long, when his turn comes he'll drop the 
> packets as the timestamp on them is too old and sound quality 
> will start decreasing ?
> If this is the case, isn't the important measure not "how 
> many are waiting" but "how long are they waiting" ? Since the 
> upper bound to load should be "low enough so they don't have 
> to drop" .
> 
> (as a fast reply I can say that I made some calls while my 
> dual 3.0 Ghz was under load 5, and they sounded good, alaw no 
> transcoding) _______________________________________________

#####################################################

What is the relation between I/O wait and load average?
by Jeff Layton

Linux follows the standard of traditional UNIX and computes its load average
as the average number of runnable or running processes (R state), and the
number of processes in un-interruptible sleep (D state) over the specified
interval.

Some other operating systems calculate their load averages simply by looking
at processes in R state. On those systems, load average is synonymous with
the run queue -- high load averages mean that the box is CPU bound.

This is not the case with Linux. On Linux the load average is a measurement
of the amount of "work" being done by the machine (without being specific as
to what that work is). This "work" could reflect a CPU intensive application
(compiling a program or encrypting a file), or something I/O intensive
(copying a file from disk to disk, or doing a database full table scan), or
a combination of the two.


Mimmus