RE: Performance with multiple users during transformat

I'd say only the first transformation shouldn't be used in measurements,
according to my current tests with Saxon and Saxon.NET, but that's Windows
specific :-)

Bhupendra, your best option (but of course not the only one) would be IMO
using a cached stylesheet. Making sure that the parsing and loading only
occurs one time will save you a ton afterwards. I have done this for quite a
few applications now, and in scenarios where the same stylesheets were used,
I've gained a considerable increase of the amount of RPM (Requests Per
Minute).

However, doing such tests would require a bit more than trying to launch the
same request from different machines at the same time, once again, talking
about Windows (Web) development, I mostly use ACT to put stress on web sites
but we've also developed some specific stress testers ourselves where ACT
couldn't provide us with enough 'app-stress'.

If other CPUs are not responding then all the processes are worked on on
only one thread, or simply there's no support for it from the application
point of view. Creating different threads could help here also.

HTH,
<prs/>

-----Original Message-----
From: Michael Kay [mailto:mike@xxxxxxxxxxxx] 
Sent: Wednesday, December 01, 2004 4:17 PM
To: xsl-list@xxxxxxxxxxxxxxxxxxxxxx
Subject: RE:  Performance with multiple users during transformation

> 
> > Bhupendra, have you monitored the load and cpu usage during the 
> > request handling?
> 
> The memeory usage never crosses 4 GB(out of 32 GB ) and all the 8 CPUs 
> also are more than 70% idle.
> 

Do remember that the Java VM takes some time to warm up. Results for the
first couple of transformations aren't representative.

Michael Kay
http://www.saxonica.com/