[SunHELP] actual, not average CPU load?

Valery Yankin sosiski at gmail.com
Tue Mar 7 01:21:59 CST 2006

   James Fogg wrote:

Speaking from my experience, you are asking for a miracle.
There are good performance tools available, but anything
which gave you the kind of accuracy and timing you are asking
for would carry so much overhead that the tool itself would
distort your information to the point of uselessness.
Hopefully others have better options/opinions for you.

I'd agree with this, and add the following -

Average CPU is a fine measure unless you are working with a real-time
system. Since Solaris/SPARC is stable up to 100% CPU load, if a
momentary resource hog consumes the system it doesn't matter. The
average system response time is going to be reflected by average CPU
load. Human users aren't very sensitive to brief load spikes. If you
*are* dealing with something that is very response-sensitive then Unix
is probably not the OS you should be working with.

Thanks, James. Yor point is clear. I would like to explain why the need for suc
h precise measurements. My aim is to monitor a process that receives data every
 15 minutes and processes it for some short time. I need then to draw a graph t
hat would show how long the process is busy processing the data. On my present 
graphs based on the results from `ps` I have something that resembles a cardiog
ram. So, people are asking me "Hey, why is that process so reluctant to free th
e CPU after the processing is over? Because on the CPU load graph we see a peak
 rise, then the line goes flat for some time (processing occurs) and then it sl
owly goes down instead of dropping. Why?" So I guess I need to make it clear fo
r the colleagues that this is the way `ps` measures the CPU load time.

The problem is, I am using sort of a factory-tailored Solaris installation with
 a fixed set of tools, and it is forbidden to install anything else. So, no pkg
_add for me.


More information about the SunHELP mailing list