Looking for framework for performance testing for derby..

[Sunitha Kambhampati <ks...@gmail.com>]

Currently we do not have performance tests for derby (except for the 2 
tests in ..tests/perf).    It will be nice addition to have performance 
tests checked in, that anyone can run.

I'm interested in looking at this - to add some kind of  framework where 
we can have performance tests for derby.   The goal is to have a 
framework where it will be easy for anyone  to add performance tests, 
run them easily, compare results on their machines.    This will help us 
to spot performance regressions and also give us a better indication of 
areas that we can improve derby.  If tests are easily available, more 
chance that others can try it out on their favorite platform/jvm 
combination and report to the list or wiki.

 From a quick google search-- Apache JMeter looks interesting.. , and so 
does JunitPerf (this has BSD license). 
Among opensource benchmarks, there is http://polepos.sourceforge.net/ 
...  They have run tests with derby. Need to look more to see how 
suitable they are for derby..
Are there already available frameworks  that we could build our 
performance tests upon....Maybe other apache projects might already have 
something for performance tests that is working for them.  Has anyone 
looked at how other projects handle it..  or have any 
suggestions/pointers, I'd appreciate it.

Thanks much,
Sunitha.

Re: Looking for framework for performance testing for derby..

[Daniel John Debrunner <dj...@apache.org>]

Sunitha Kambhampati wrote:
> Currently we do not have performance tests for derby (except for the 2 
> tests in ..tests/perf).    It will be nice addition to have performance 
> tests checked in, that anyone can run.

+1

I think it would be good if the performance tests could utilize the 
JUnit work, so that, for example, getting connections is handled in a 
single way, allowing the performance tests to run with J2ME, 
DriverManager, DataSource, single leg of an XA transaction etc.

If the performance test was also a Junit test then they possibly could 
serve double duty as a function test. Always good to know that commits 
did not break the functionality that performance tests require. :-)

Dan.