Validation of HTTP Response Time From Network Traffic as an Alternative to Web Browser Instrumentation

The measurement of response time in hypertext transfer protocol (HTTP) requests is the most basic proxy measurement method for evaluating Web browsing quality. It is used in the research literature and in application performance measurement instruments. During the development of a website, response time is obtained from in-browser measurements. After the website has been deployed, network traffic is used to continuously monitor activity, and the measurement data are used for service management and planning. In this study, we evaluate the accuracy of the measurements obtained from network traffic by comparing them with the in-browser measurement of resource load time. We evaluate the response times for encrypted and clear-text requests in an emulated network environment, in a laboratory deployment equivalent to a data centre network, and accessing popular Web sites on the public Internet. The accuracy for response time measurements obtained from network traffic is noticeable higher for Internet long distance paths than for low-delay paths (below 20 ms round-trip). The overhead of traffic encryption in secure HTTP requests has a negative effect on measurement accuracy, and we find relative measurement errors higher than 70% when using network traffic to infer HTTP response times compared with in-browser measurements.