Can anyone out there share their experiences with the super user concept when load testing? It's a very interesting theory, as the load test user does not have a think time as humans do. Here is a good article on perf testing that talks about the super user and a formula as well:
One other thing... per the super user formula, I need the metric 'requests/page'... How do I find or calculate this? I am using Perfmon.. and can't find this. I'm also using TestComplete for load testing and they don't have this metric either... so I have a question into them on a formula as well.
Maybe I'm using the wrong terminology here... super user... non-human user [img]/images/graemlins/smile.gif[/img] ... any other formulas that people have used for this? Out tool does not incorporate think time.. so our 'users' are super users. Any feedback/discussion on this would be mucho appreciated.
There's benefits and drawbacks with user this super user concept. If you're looking more at throughput versus end user simulation, by all means cut out the think times and crank up those transaction rates.
One drawback I see immediately is session management. If you have 300 concurrent users on a system and test with 300 users with think times, etc. it's not the same test if you ran 20 super users with no think times. Your throughput might be the same, but they are very different tests and could result in very different system behavior.
Thx for the thoughts. I agree.. I mean, ideally, it would be nice to run tests with and without the think time and compare the 2 to see how close to the mark we are on our ratio of super users to real users. Unfortunatley, our current loadtesting tool does not incorporate think times. This seems to be a big limitation... however, even with think times, nothing is perfect (or truly a real representation).
For our products, our requirements will maily consist of specific steps that a user must to perform to get to an end result (in other words, the user can't perform step 3 without first performing step 2.). There are authentication issues as well (user MUST login to do anything in the app), so throughput does not appear to be the best possible type of load.
For throughput, we would most likey load via web services.