Transport Protocols for Next-Generation Networks and Applications

Today's Internet is very different from what it was intended to be forty years back. Its capacity has increased by orders of magnitude and the last mile access medium has transitioned from wired to wireless communication technologies. Yet the architecture of the Internet has not fundamentally changed, the same Internet protocol stack remains the vital core of the network. The assumptions made in the development of the Internet protocols were based on wired links and a limited set of applications. Most of these assumptions do not hold for the new types of applications running on wireless links.

The layered-siloed architecture of the Internet has maintained a high level of abstraction that made it possible to adapt to new link types (such as optical networks, WiFi, cellular, satellite, etc.) and support new applications (such as realtime multimedia communication and video streaming). As cellular and other wireles networks take their place as the ubiquitous link layers of the future, wireless resources will become scarcer than ever. On the software side of the world, it is crucial to make the best use of the available resources. Unfortunately, combining the new applications and link layers with the current architecture results in poor application performance and inefficient link layer management.

In this dissertation, we focus on the transport layer which provides the abstraction interface between the link and the application. We argue that a fresh look at the existing transport layer solutions is necessary to fully utilize the capabilities of emerging link-layer technologies and enhance the services provided to new types of applications. We present two solutions to address the interaction of the transport layer with both the link and application layers. Our first solution, Link-Coupled TCP, uses explicit cross-layer communication to allow applications to explicitly configure the desired trade-off between link utilization and queueing delay. By leveraging the architecture of emerging 5th Generation (5G) networks, it allows concurrent flows of applications with heterogeneous requirements to coexist without interfering with one another and without lowering the overall utilization of the wireless link. Our second solution, Application-Aware TCP, uses implicit cross-layer communication to identify the type of traffic generated by the application. It adjusts the congestion control parameters of the transport layer in order to optimize the user's quality of experience. Application-Aware TCP improves the performance of web browsing and adaptive video streaming when competing with bulk transfer traffic without negatively impacting bulk transfer traffic on the long run.

Evaluating new solutions in realistic scenarios requires reliable tools that cover the different components and layers of the network. Network emulation provides enough flexibility by implementing a fully contained software replica of one of the network layers which can be tuned to act like its real counterpart under different conditions. For the scope of our solutions, we focus on three components: the wireless devices, the physical RF medium, and the wired portion of the network. We propose a set of guidelines for performing reliable experiments using the existing network emulation tools in Linux. We also present a wireless network testbed for accurately and reliably emulating the physical RF medium.