Setting Realistic Think Times in Performance Testing: A Practitioner's Approach.

Think time in performance testing is the time spent by users between their interactions with a software system such as viewing responses, navigating on the screen, entering data etc. The value of think time has a significant effect on the results of the performance test in terms of the system load which is generated during the test. However, the use of realistic think time values has not been explored in detail either by researchers or performance testing practitioners in IT industry. Practitioners either use subjective think time values or values generated using uniform distribution. Our work provides an algorithm for extracting think time values by processing web server logs. The steps for using the extracted think time values are described for load testing tools such as Rational Performance Tester, JMeter and LoadRunner. The algorithm is used to extract think times from web server logs of a software system servicing a large number of users. The real-world think times are compared with the think times being used in the performance testing of the application. This work will help practitioners in the IT industry to generate more accurate and reliable performance testing results by moving away from the use of experience or intuition based think times to incorporating representative user delays in their tests. The proposed approach requires minimal effort and can be seamlessly made part of the performance testing process.