Effective analysis of secure web response time

Abstract

The measurement of response time in web based applications is a common task for the evaluation of service responsiveness and the detection of network or server problems. Traffic analysis is the most common strategy for obtaining response time measurements. However, when the traffic is encrypted, the analysis tools cannot provide these measurement results. In this paper we propose a methodology for measuring the response time in HTTPS traffic based on the flow of data in each direction. We have validated the tool with real traffic and with a worst case scenario created in a testbed. When pipelining is present in the encrypted HTTP 1.1 traffic, it results in a small error in the measurement (between 5% and 15% of error for the 99.9 percentile of the real response time). However, pipelining support has almost disappeared from modern web browsers; this makes the estimation provided by this methodology very accurate in real traffic measurements, even for low probability response times. More than 98.8% of the over 8.6 million request-response times we measured in our campus Internet link were obtained without any error.