What Were They Thinking: Modeling Think Times for Performance Testing

When modeling users interacting with a system, the concept of think time will arise. Besides developing a workload model and recording scripts, a think time analysis is necessary in order to have an accurate test and appropriate system throughput. Think time is the time the user spends between tasks. Example tasks include: selecting from a menu, entering data into a text field, reviewing a list of results before selecting one or more of them, and navigating to a completely new task. These tasks are more than the physical motion required to interact with the system’s user interface; they include the thinking time that causes the user to make these choices. What cannot be determined when examining production data is whether or not the user is truly interacting with the system during the time between tasks; idle times are hidden within think times. This paper details a start-to-end think time analysis. It extends the analysis that appears in [Wil10a]. The analysis uses real production transaction data to derive think times. After processing these data, the analysis constructs a think time model for use in a load generator. Then, the analysis attempts to refine the model to provide more accurate times for given situations. The more accurate the model, the more confidence there will be in the analysis of the performance test. Finally, an operations analysis is performed as a check. A proprietary transaction system that supports logistics and maintenance for military equipment provides example data. For the purpose of anonymity, most system details are omitted. We do not need to understand the system in order to analyze the think time data. Ultimately, the think time data will be fed into a performance test for the next generation system.