Many thanks FrustratedUser. This was the right answer. However I didn't succeed in making it work. I have one question. On the window "Workflow - Calculate the Virtual Users" (Adjust Workload -> Workload Wizard -> Next), with Static or Dynamic selected for the "Pace of transaction think times", is it possible to set a Global Session (Time) lower than the Baseline Session (Time) ? I would just like to see the think time adjusted in order to meet the transaction rate I want. Can I do that only in a demo (not in real conditions with server slowdown under heavy load) ?
After several tests, I think that the throughput is the number of sessions per hour (the "Sessions" box in the "Workflow - Calculate the Virtual Users" window). However, Silkperformer always reaches a slightly lesser throughput
No matter the tool or your goal, it's almost impossible to consistently and accurately reach whatever target you set EXACTLY. No matter the target the team has determined in the applications I test, be it throughput, HTTP hits/sec, transaction/hour or whatever, I consider success as being +/- 10% of that goal. I strongly recommend you do the same. The one and only constant of performance testing is that the large amount of variables during the testing process negates consistently achieving a higher degree of accuracy in the results.
There are a few things you could do to make each test closely resemble the last. For instance, don't randomize anything, only use static values, have hard-coded thinktimes, and you can use
<font class="small">Code:</font><hr /><pre>GlobalWaitFor("StepName",ALL_USERS);</pre><hr />
for each and every action. I wouldn't recommend any of that though. Hope this helps.
Thank you for that answer. I can add that the true goal of a test is to simulate real conditions, and randomization helps to do that. Thanks again because you just gave me materials to argue in favor of "a successful test"