performance tuning at process level

[Lars Tunkrans]

j wrote:

> Hi all,
>
> I am developing an application on a solaris with config as follow :
> SunOS HKTA2SN02 5.7 Generic_106541-24 sun4u sparc SUNW,Ultra-60
> System = SunOS
> Release = 5.7
> KernelID = Generic_106541-24
> Machine = sun4u
> Origin# = 1
> NumCPU = 2
>
> After runing the application, we run :
> sar -u 1 100
> ps -ef -o pcpu, args | sort -k 1
> uptime
> sar shows that the idle percentage is 0 (which I understand the cpu is 100%
> used) and %usr is about 80. However, in ps, I find that only a summation of
> 40 is found. So we don't know where does the rest goes. Uptime shows load
> average of as high as 6.7.
>
> What can I do pick out the process which eat the cpu % ? I can only use
> command come with solaris and cannot install new one (like top).
>
> For your reference, my program is a cgi application as link checker which
> parse the html, find out the link and open socket to connect to external
> server to see if the page is there, or if the page is on local server, it
> will check if the exists physcially (no socket opens here). Please advise if
> you think where the hotspot is.
>
> Thanks
>
> Perseus
>

Dont run Sar every second !
The result you get will be wrong !
The result will include to much of sar itself !

Sar collect mean values over longer periods of time.

If you use less than 10 seconds as the sample time
the result will be tainted.

//Lars

[Darren Dunham]

j <perseus_medusa@hotmail.com> wrote:

> After runing the application, we run :
> sar -u 1 100
> ps -ef -o pcpu, args | sort -k 1
> uptime
> sar shows that the idle percentage is 0 (which I understand the cpu is 100%
> used) and %usr is about 80. However, in ps, I find that only a summation of
> 40 is found. So we don't know where does the rest goes.

Which PS fields do you use? Most of them are "time averages" with a
time constant near a minute. If you run a single CPU process (which
takes appx. 100% cpu) and check PS quickly, it will not be near 100%.

Note this process on an otherwise idle box...

% perl -e 'while(1){$i++;}' &
[1] 518
% ps -o pcpu -p 518
%CPU
8.8
% ps -o pcpu -p 518
%CPU
27.3

It will eventually reach near 100, but that means that taking a single
'ps' doesn't necessarily give you numbers that you can sum.

> Uptime shows load
> average of as high as 6.7.

> What can I do pick out the process which eat the cpu % ? I can only
> use command come with solaris and cannot install new one (like top).

'prstat' and 'prstat -m'. The second one does not use the time-average
values.

--
Darren Dunham [email]ddunham@taos.com[/email]
Unix System Administrator Taos - The SysAdmin Company
Got some Dr Pepper? San Francisco, CA bay area
< This line left intentionally blank to confuse you. >