Passive Analysis of Web Traffic Characteristics for Estimating Quality of Experience

Abstract

Detecting user's quality of experience is a method which deals with estimating how a user perceives the network performance. Although user's behavior is very complex, only analyzing a large amount of operational network data enables us to effectively detect the users satisfaction from the network. One important parameter to detect user's behavior is user's interruption of the TCP connections in case of bad performance. In this paper we take passive network traffic measurements at the TCP flow level to monitor the effect of packet loss on Web traffic characteristics, such as TCP connection duration, size, and inter-arrival time of connections. From these traffic characteristics correlation with end user behavior (interruptions) is analyzed. Different trends are identified in the duration and size of connections with increasing loss rates. We explain and confirm the reasons for the different trends with different parameters. We discuss whether the end user behavior, such as interrupting connections, be identified from these traffic characteristics. We also want to know whether it provides us reasons for a connection being small or large, both in duration and size, due to user's behavior. A precise way of estimating packet losses for passive measurements is devised, giving a lower bound on the total packet losses in a TCP connection.