Quantification and prediction of end-user perceived web-browsing quality

Web browsing is one of the most popular data services over both fixed and mobile networks. The end-user perceived quality of web browsing depends on the entire end-to-end chain: network performance (server load, number of servers, link capacity, etc.), the user equipment, and the application performance (the browser itself). To evaluate the web-browsing quality, a model is formulated based on a novel approach that relates the measured response and download times to the perceived web- browsing quality. To this end, a number of web-browsing quality experiments was set up that allow to manipulate the response and download times of a small web browsing session. In the experiments, a subject repeatedly retrieved the same two web pages, one search page and a page that results from the requested search. The response and download times of the two pages are manipulated in three different experiments, representing slow, medium and fast Internet access. Major results are that the perceived quality is inversely proportional with the logarithm of the (weighted) session time and that the last download time dominates the perceived quality. This last download effect is stronger as the maximum expected session time in an experiment gets shorter and is thus especially important for fast internet access. With the optimal weighting, the correlation between objective timing measurements and perceived quality is above 0.93 for all experiments that were carried out. Without weighting of the response and download times, the correlation for the fast link experiment dropped to a value as low as 0.72.