Siegfried, > +) Canoo WebTest is not a load testing tool - at the end of the day it > does do too much to produce high load easily
not that much: only reporting is too much, but this can (or could be disabled) > +) load testing tools are not functional test tools because it would > make them slow - and don't forget the reporting part of load testing tools I'm convinced that WebTest (or something else based on HtmlUnit) can open a new way of "functional load testing" because HtmlUnit is fast enough for that (users already reported that they run successfully a few hundreds of WebClient instances on desktop computers to perform load testing of AJAX apps). The reporting part of load testing tools is very important, that's why something like a JMeter Sampler for WebTest would make sense. This would surely not allow to simulate as much "users" per single computer, but enough to be interesting. > +) I do agree that there is some overlap but you are looking at two > different type of tools. > > So these are things I do > > +) create a Canoo WebTest covering most of the application > +) write a simple JMeter test to capture the the common use cases with > minimal verification > +) fine tune the JMeter test to create realistic load (this is the hard > part - creating test data, ensure that the tests executed reflect the > current/anticipated user behaviour - this sometimes involves creating > millions of database records and analyzing gigs of access logs) > +) run the JMeter tests in parallel to the Canoo webtests such an approach is fine... until you have too much logic on the client side. And due to AJAX, this is the trend. In this case, recording the http traffic doesn't allow to reproduce load that would represent the "real life" of the application. > So i can gather my performance data and still be sure that the > functional tests pass when running high load (the maven plugins have a > "loop" command to execute the webtest until the first problem is > encountered). At the end of the day I try to combine the sweet spots of > a functional and load test tool ... Load testing is surely not only running a few tests with high concurrency. But when I see how load testing experts deal with very low level code, I'm quite sure that it is not the most efficient way to write maintainable code that reflects the application's state and really think that WebTest can bring something new in this area. Cheers, Marc. -- Blog: http://mguillem.wordpress.com _______________________________________________ WebTest mailing list [email protected] http://lists.canoo.com/mailman/listinfo/webtest
