Improving the Adaptability of the End-host : Service-aware Network Stack Tuning

Abstract: The Internet of today is very different from how it used to be. Modern networked applications are becoming increasingly diverse. Consequently, a variety of requirements must be met by the network. Efforts to make the underlying mechanisms of the Internet more flexible have therefore been made to adapt to this diversification. In this thesis, we explore how information about application requirements can be leveraged to optimize the network protocol stack of end-hosts during run-time. In addition, we improve the visibility of the network to the end-host in order to enable additional flexibility in the usage of the network's resources.We conduct tests in real-world testbeds and examine how services might be developed to optimize latency, throughput, and availability for various network traffic scenarios, including 360-degree video streaming, drone autopilots, and connected vehicles. We show how multi-connectivity, where the end-host is connected via multiple network paths simultaneously, may be used to significantly reduce latency and increase availability, while minimizing the overhead imposed on the network by carefully considering the network selection process. Furthermore, we describe an architecture that allows the user equipment and network functionality inside the 5G core network to cooperatively optimize the resource usage of the network.