الفهرس 
Chapter 1 IntroductionOnly 14 pages are availabe for public view
 |  | from | 137 |  |  |
| | | from | 137 | | |
Abstract
Performance testing increases client’s confident level and ensures that the Application Under Test (AUT) behavior is as expected under extreme conditions. In our research work, we are focusing on both load testing and stress testing of web applications. Generating the test load is relatively straightforward, but there is a tactic in creating a realistic load and increasing that load in a controlled way. This is to enable monitoring the performance of the application and analyzing the test results. In our research work, we study how performing realistic load of web application leads to realistic results. We design three controlled and creative test phases. In the first phase, we define a realistic scenario between client and server while taking into account the users think time. We also define a number of virtual users and control the load of virtual users by using a new methodology. The proposed methodology increases the number of users step by step, uses ramp-up algorithm, holds the test for some period of time, and then uses a ramp down algorithm to decrease the number of users steeply. In the second phase, we run the test. Once the test runs the users begin sending requests to the web applications according to our building scenario during that the results are automatically recorded. Finally, in the third phase we analyze the recorded results using machine learning techniques. We run experiments using one machine, two machines, and three machines, and compare the results of these experiments. All experimental results show that our proposed methodology allows the load to be increased slowly so that the performance could be measured at different load levels. It also shows that our proposed methodology is robust and could be efficiently used to detect performance bottlenecks. We also use it to classify the response codes to detect the errors and measure their percentages if exit as well. We also provide some solutions to decrease the error percentages.