Since its inception in the early 70’s, the Internet has expanded to interconnect billions of computers. The last decade has seen the rise of multi-connected devices, such as smartphones that can attach to both Wi-Fi and cellular networks. Endhosts can leverage this availability of multiple network paths in two major ways. First, they can aggregate the bandwidth of the available networks. Second, they can support seamless network handover when one of the network paths fails. Applications rely on transport protocols to carry data between endhosts. These applications can have various requirements. While simple bulk transfers want to complete as fast as possible, interactive applications (such as voice-activated ones) have strict latency constraints, especially under mobile conditions. Furthermore, the device user might enforce additional constraints, such as limiting the cellular usage on smartphone for both monetary and energy purposes.
This thesis explores how multipath transport protocols can adapt to their serving use cases. First, we quantify the performance of the state-of-the-art Multipath TCP on smartphones by performing two measurement campaigns. Second, based on our previous findings, we tune Multipath TCP for the smartphone use case and evaluate it with real Android phones. Next, we focus on the emerging QUIC protocol, design two versions for its Multipath extensions and evaluate their benefits in various scenarios. Finally, we reconsider the architecture of transport protocols such as their operations can be finely customized on a per-connection basis thanks to the dynamic injection of plugins.
Jury members :
Podcast of this thesis is available here