On the Use of TCP Interruptions to Assess User Experience on Web

The Quality of Experience (QoE) is becoming anincreasingly popular research area among the ICT relatedindustry and academia. While the major work was beingdone on the Quality of Service (QoS), the past decade haswitnessed a vacuum in the area of QoE because QoE is rathera new notion. This is quite evident from the inconsistentdefinitions of QoE that we find in the literature. However,the area is now gaining more attention and maturity as thereis increasing realization that ’the better the QoS’ doesn’tnecessarily mean ’the more happy the users’. It is possibleto have excellent QoS but poor user experience [1]. Forservice providers to compete in the market, they need to givemore attention to how user perceives the service than justworking on QoS. This further supports the argument thatthere is more than just technology-centric approach when itcomes to satisfaction of users. We often observe that newapplications and services that claim additional performanceare rejected by the users. Consequently, there is a need ofa user-centric approach for the design and monitoring ofnetwork applications and services.It stimulates the need for an online passive mechanism tomonitor the user activity in real-time. The mechanism mayvary with the type of application. In this work, we keepour focus on Web browsing application as it is one of themost popular applications on the Internet. The aim is to letthe users browse freely and monitor their activity instead ofasking them subjectively about their usage experience.Monitoring of the TCP connection interruptions could beone of the ways to have indications about users’ feelingson the Web. This is based on the idea that, in case of badperformance, users break a Web browsing session by pressinga reset or stop button in the browser hence generating aTCP reset on TCP flow level. However, several other causesmay explain the generation of TCP resets as described in[2]. In this work, we apply the TCP connection interruptioncriterion presented in [3] to filter the TCP resets generatedby the users. We apply this criterion on the Web traffic of anoperational mobile network.In our previous work [4], we correlated the user sessionvolumes with the network performance i.e. throughput andpacket loss. We found that the volumes increase with theincreasing throughput implying that the happy users surf morein the case of better performance. In this present work, wecarry this analysis further to observe the user interruptions inrelation to transfer sizes and durations. We want to see howthe users react to bad performance and try to explain whywe observe smaller mean transfer sizes. Do the users avoidlaunching large transfers in the case of bad performance ordo they try to get the same files but stop the transfers thatare becoming too long?II. M