Download files
Abstract
Due to the proliferation of data-hungry applications in smartphones and other wearable smart devices, demand for wireless data bandwidth is increasing rapidly. Consequently, researchers are exploring new schemes and frameworks to satisfy this demand. Multipath TCP (MPTCP) is proposed to not only enhance end users’ throughput but also to increase the resilience against link failures. MPTCP is a new TCP extension that can leverage availability of access to more than one network and transfer traffic along several network paths. Smartphone application are thus one of the leading beneficiaries of MPTCP. However, MPTCP impacts battery life, which is a perennial problem for smartphone users. Furthermore, it can introduce security threats in the form of covert channels. In this thesis, we explore these two aspects of MPTCP, namely 1) how can we make MPTCP more energy efficient for mobile users, and 2) how can detect and prevent the security threats covert by channels.
In order to address MPTCP energy management in mobile devices, we developed several Markovian optimisation schemes. We showed that it is possible to save energy by accounting for different interface signal qualities and transferring file sizes. Additionally, we determined that MPTCP can make LTE more energy efficient by cutting its tail energy. We extended our schemes using game theory and Q-learning to address the effect of other decision makers on the performance of MPTCP. In doing so, we showed that it is possible to increase the energy efficiency of MPTCP by having no or partial information about other players in the environment. Therefore, we proposed an initial path selection algorithm by employing a Bayesian game to augment MPTCP. Furthermore, we showed that, currently, without these optimisations, MPTCP is not an optimal choice for energy efficient transmission compared to using TCP with WiFi or LTE, as any upload, such as an acknowledgement, consumes a great deal of a mobile device’s energy.
We also showed that MPTCP has potential for introducing additional storage and timing covert channels. Moreover, detection with full or partial access to all of the subflows poses different kind of challenges and difficult to detect because of distinct MPTCP reactions to network parameters. Finally we show that covert storage channels can be detected and prevented by statistical tests and traffic normalisation.