Non-Intrusive Network-Based Estimation of Web Quality of Experience Indicators

Abstract: Quality of Experience (QoE) deals with the acceptance of a service quality by the users and has evolved significantly as an important concept over the past 10 years. Network operators and service providers have gained interest in QoE-aware management of networks, in order to better fulfill end-user demands and gain a competitive edge in the market. While this growth promises new business opportunities, it also presents several challenges to the networking researchers, which are mainly related to the assessment of user experience. Several QoE assessment models have been proposed to estimate the user satisfaction for a given service quality. Most of them are intrusive and require knowledge of the content reference. In contrast, the network operators require non-intrusive methods, which allow models to be implementable on the network-level without having much knowledge about that reference. The methods should be able to monitor QoE passively in real-time, based on the information readily available on network level. This thesis investigates indicators, which are intended to be used in the development of non-intrusive network-based methods for the real-time QoE assessment and monitoring. First, a bridge is made between the user and the network perspectives by correlating the user traffic characteristics measured on an operational network and user subjective experience tested on an experimental platform. It is shown that the user session volume appears to be an indicator of users’ interest in the service. Second, the TCP connection interruptions are investigated as an indicator to infer the user experience. It is found out that the request-level performance metrics show stronger correlations between the interruption rates and the network Quality of Service (QoS). Third, a wavelet-based criterion is devised to assist in the identification of those traffic gaps, which may result in the degradation of QoE. It can be implemented on the network-level in quasi-real-time to quickly identify the user-perceived performance issues.