Home clients can use their access to the Internet for different purposes such as file sharing via P2P applications, gaming, or Web browsing; the last one is the focus of this work.
When browsing the Web, the time elapsed between the click on a URL and the rendering of the Web page, referred to as page load time, is the key performance metric. When the
page load time is higher than a few seconds, the user experience suffers significantly. We have developed a three-tier system that (i) captures in the browser the events necessary
to measure the page load time (ii) captures at the network access all incoming and outgoing packets, and (iii) correlates the measurements made at different machines. The capture
at packet level allows us to compute the contribution of the various steps that affect the page load time such as DNS resolution, server response time, data transfer time. Correlating the observations made at different machines that share a major part of the network elements can help identifying the root causes for high page load times. We will present the architecture of our system and some examples that illustrate its use.
