You may also want to look at how various actions effect performance. D.
On Wed, Dec 9, 2009 at 11:02 PM, Markus Kohler <[email protected]> wrote: > Hi all, > > Here are my thoughts about how I want to proceed with the performance test. > > Any comments are welcome! > > *Single User tests* > > So far I didn't put any real load on the esme server, but just ran tests > with one single user. BTW, this is how the ABAP performance experts do it. > They rarely do tests with many users :-) > > I used Selenium to automate the browser, which turned out to be pretty easy > and we could still find some issues. > > There are a few things I intend to test this way. So far no user follows any > other user. I will write a test that does that, which in itself might be > interesting, but it would also be interesting of course to see later on > whether that influences the peformance of message sending. > > *Load ( multi user) Tests* > > Load test, simulate the load of many users, and are typically done using > HTTP record and replay tools (Loadrunner, JMeter etc). Those tools are > pretty dumb, they don't really emulate the browser. So no javascript is run > and no rendering is executed. They just send HTTP requests and slightly > modify them for each user. There advantage is that they are very efficient. > > > For web applications that use Comet, such as ESME, this is not really an > option. > > Browsermob is an interesting load testing service running in the cloud that > uses real browsers for load testing. They combine it with the HTTP request > replay method. They differentiate between real users (browser based) and > virtual users (http replay based). > > The difference between real users and virtual users is explained here > http://blog.browsermob.com/2009/12/faq-whats-the-difference-between-a-real-browser-user-and-a-virtual-user/ > > It seems browsermob has a free testing period. I might use that for some > tests, but it' s not the ultimate solution because I can't really profile on > STAX. > > Selenium can also drive tests using HTMLunit, which simulates a browser in > Java including Javascript, but without rendering. > > Plan A is to try whether this works. If it doesn't work, > > Plan B would be to combine the single user Selenium tests with tests against > the "REST"-API. > > Regards, > > Markus >
