On this week I participated in the course Introduction To Load Testing by Kristo Kuusküll. Here are some interesting notes (not quotations) from this course. I emphasize that I am not a load tester, so this area is new for me.
* It is necessary to know the frequency of clicking and typing of end-users. If client can not define these requirements you can take this information from live logs.
* Load testing software does not simulate the browser.
* Load testing takes a couple of days even among professionals.
* You need tests and sometimes even a load tests for load tests.
* Before overloading the application always run the tests for 1 user (thread). If it is ok only then you can run it for 1000.
* Load testing software and the application never should work on the same machine, because load testing software increasing the CPU%.
* If CPU usage is grater than 75%, then you can not trust the results.
* It is necessary during the testing to browse the application in the browser - only that you can see the whole picture.
* You should check not only test status (fail\ok) but the response data itself. There might be successful response (code 200) with text of error (sorry, the server is overloaded, please contact the administrator).
* The best software for beginning is JMeter. But it takes a lot of memory, so you can not use it for sizable projects (more than 300 users).
* In JMeter different reports take different amount of memory, so you can increase the number of threads by disabling some result reports (Summary Report is the best).