The new web-based applications and services, which are becoming more and more popular every day, have completely changed the way users interact with the Web. In less than half a decade the role of users has changed from passive consumers of information to active and dynamic contributors to the contents offered. Moreover, this trend is expected to rise in the incoming Web.

This user's behavior is a major concern when defining web workloads in order to precisely estimate system performance for the current Web.However, the intrinsic difficulty to represent the user's dynamism in a workload model has led many research works to still use workloads non representative of the current web navigations.

This dissertation focuses on characterizing and reproducing more realistic workload for web performance by mimicking the behavior of the real web users. 

The state-of-the-art in modeling and generating workloads for web performance studies presents several lacks in models and software that represent the different levels of user's dynamism.This fact motivates us to propose a more accurate workload model and to develop a new workload generator based on this model. Both of them have been validated against a traditional workload generation approach. To this end, a new testbed with the ability of reproducing traditional and dynamic workloads has been developed by integrating the proposed generator with a commonly used benchmark.

In this Ph.D dissertation we also analyze and measure for the first time, to the best of our knowledge, the impact of using representative dynamic user workloads on web performance metrics instead of traditional workloads. Experimental results demonstrate that the use of an accurate workload model that considers user’s dynamism when navigating the Web clearly affects system performance metrics as well as the stress borderline of the server.

Finally, we explore the effect of considering the \acl{UBI} as a part of user’s dynamic behavior on web workload characterization.The study proves that representing user's dynamic interactions with the provided contents allows users to achieve their navigation goals sooner thus increasing the productivity of their navigations. In addition results demonstrate that this type of navigations also affects the stress borderline of the server and system resources utilization.