On Friday 22 March 2002 09:36 am, Luke Opperman wrote:
> Hi again -
>
> So I just watched Webware go on a sustained (30second?)
> 100% CPU burst. Which led me to wondering how I could go
> about profiling what servlets are responsible, or if it's
> just a traffic thing, or... graphs are pretty, how feasible
> might it be to provide stats to RRD or similar? I'm looking
> for both overall AppServer health/trends, and per-Servlet
> stats. Maybe I can automate something from the Activity
> log..
>
> Thanks for any thoughts, experiences, etc

I've gotten a lot of mileage out of Python's own profile module. But 
I've only used it on command line utils that terminate. Haven't tried 
it on WebKit apps yet.

Also, it slows down your programs quite a bit. Everyone says 10 X, but 
my experience is 3-5 X. The point is that you probably can't run 
profiling on your production server (due to the slowdown), which then 
means you have to find a way to simulate the same loads and inputs from 
production on your test box (in order to get the most realistic 
information).

I use Webware/bin/printprof.py to dump the stats that profile 
generates. No point in writing that snippet over and over.

Using the profile information, I've gotten code to run anywhere from 
1.3 to 5 times faster by attacking the parts of a program that consume 
the most time.

Don't forget to re-profile after each change in code, as the profile 
can change quite a bit as you do this.


If you try this on an app server, I would be very interested to know 
your results, tips, etc.

-Chuck

_______________________________________________
Webware-discuss mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/webware-discuss

Reply via email to