Using Log Records for Web Application Load Profiling

Abstract

The performance of Web applications is important for efficient operation and is usually addressed by testing with appropriate performance testing tools. Typically, expected loads are simulated to evaluate system performance and stability. In this paper, we present a new method to support load profiling that improves load planning of web applications by analyzing actual user behavior. Based on server log data and machine learning algorithms, we identify user patterns that allow us to more accurately adapt the application to real-world needs. This retrospective testing approach allows for more realistic simulation of user actions and optimization of resource usage. Combining expert assumptions with analysis of actual usage leads to more robust infrastructure design and optimization, improving application performance under realistic conditions. The methodology developed to support log-based load profiling of Web applications enables the collection, cleansing, and preparation of log data and the identification of user groups and their behavior.