Web performance analysis based on user experiences

 The thesis is composed by three parts: performance analysis, performance diagnosis and a new methodology for critical path analysis. The first part intro duces our initial platform which is based on browser-level measurements. We explain measurement metrics that can be easily acquired from the browser and indicators for end-user experience. Then, we use clustering techniques to correlate higher-level performance metrics with lower level metrics. As we will show, such clustering methods can be used in different scenarios to study the relationship between QoE and QoS, and compare and explain the performances experienced in different homes.

 

In the second part, we present our diagnosis tool called FireLog. We first discuss different possible causes that can prevent a Web page to achieve fast rendering; then, we describe details of the tool's components and its measurements. Based on the measured metrics, we illustrate our model for the performance diagnosis in an automatic fashion. Finally, we use both controlled experiments and long-term wild deployments to evaluate FireLog.

 

In the last part, we propose a new methodology named Critical Path Method for the Web performance analysis. We first explain details about Web browser's intri nsic features during page rendering and then we formalize our the methodology. By adopting this critical path method, we transform Web page rendering process into a Directed Acyclic Graph. With help of the assigned timings, we can derive a final critical path that has a direct impact on the ove rall Web page rendering performance.