OK. I ran some basic tests on one of my machines. Just for basic info
it is a P4 2.5 GHz with 1 GB of memory running RHEL 3.
The only thing I did was set up JMeter to login to the portal as user
cocoon. In all the tests the computer was maxed at 100% cpu.
Before the change:
5 threads login repeated 10 times: Avg 3.4 seconds, Max 27 seconds.
10 threads login repeaded 5 times: Avg 6.760 seconds, Max 22 seconds
After the change:
5 threads login repeated 10 times: Avg 1.3 seconds, Max 2.6 seconds
5 threads login repeated 20 times: Avg 1.2 seconds, Max 2.5 seconds
10 threads login repeated 5 times: Avg 2.4 seconds, Max 10 seconds.
10 threads login repeated 10 times: Avg 2.1 seconds, Max 13 seconds.
The change has been checked into 2.1. I'll test it on 2.2 and check it
in also.
Ralph
Carsten Ziegeler wrote:
Castor seems to have a lot of useful little "hacks" - I just found out,
that we can prevent castor from checking for default constructors which
I really needed for 2.2 - it's there, you only have to find out how to
configure it :). Im curious how the matches configuration looks like :)
Carsten
Ralph Goers wrote:
OK. I figured out how to use the matches attribute and was able to
verify that it doesn't throw ClassNotFoundExceptions all the time. I'll
do a little load testing next to see what kind of difference it makes on
throughput.
