40 users != the same user 40 times

If you are going to do load testing it doesn't make any sense to generate a load that doesn't follow actual use. It's dumb and it'll point you in the wrong direction. Garbage in, garbage out.

We ran into that when doing some load testing with jmeter. At first we didn't know how to make our thread group use different logins so we just hard coded one login and created a bunch of threads. Not how the app will be used and performance was less than happy for me. Acceptable but not happy.

So today and yesterday I figured out 2 different ways (other than one suggested to me by Peter) to simulate different logins. The first was to add a user defined variables, add two rows (login / password) and just create more columns of users. This was okay for about 2-15 users, but as you create more the columns keep on shrinking and it's really ugly (bug). The second was to get CSVRead working (which was a bi%ch) and just have a file with user names and passwords. Much nicer and cleaner. ;-)

Aside: if anyone is having troubles with JMeter __CSVRead function, to get it working for me I had to take out the spaces between the function arguments (after the commas). *sigh*

Now with a more realistic load our app is handling a much better load and we are able to take a look at the real places in our app that are bottlenecks.

Next steps are to model different realistic situations so there are no surprises.