Programming Models and Extensive Evaluation Support for MPTCP Scheduling, Adaptation Decisions, and DASH Video Streaming

In this dissertation, we identify that the analysis, implementation, and evaluation of communication systems is hindered by two obstacles: i) missing abstractions and the resulting implementation complexity and ii) the required extensive evaluations for today's large configuration spaces and heterogeneous network environments. A prominent example is Multipath TCP - today's de facto multipathing transport protocol. Multipath TCP packet scheduler innovations are hindered by the implementation complexity of the Linux kernel network stack and the required analysis for a variety of applications and network conditions. To tackle the first obstacle, we are the first to propose a programming model as abstraction for the design and development of Multipath TCP schedulers. We introduce the ProgMP programming model, which provides a powerful specification language and a high-level API to specify executable Multipath TCP schedulers. We show the strength of ProgMP by implementing 13 novel general purpose, preference-, and application-aware schedulers tackling diverse objectives. As part of these schedulers, we propose the first redundant Multipath TCP scheduler and show that this scheduler significantly reduces latency for applications with tight latency requirements but only moderate throughput needs. We use ProgMP for a detailed analysis of design decisions for the incorporation of redundancy to balance latency and throughput. We further propose schedulers that retain fine-grained throughput or latency objectives, or improve the interaction with upper layer protocols such as HTTP/2, while preserving path preferences. Our detailed emulation-based and real world measurements show that ProgMP enables timely scheduling decisions and a wide range of executable, novel Multipath TCP schedulers. Besides ProgMP, which is our main contribution to overcome the first obstacle of missing abstractions, we further introduce programming models as abstraction for the adaptation decisions of adaptive communication systems. Here, we propose to specify the adaptation decision with event condition action rules and learn rules for a given utility function with genetic programming in extensive network experiments. Finally, we propose a programming model for the specification of topology adaptations in communication systems based on topology graph patterns. To overcome the second identified obstacle and foster extensive evaluations, we present the MACI framework for the management, scalable execution, and interactive analysis of extensive network experiments. In essence, MACI is a combination and integration of established tools to foster rigorous, seamless evaluations throughout the research process. We discuss our MACI experiences during i) the development and evaluation of our proposed ProgMP schedulers, ii) the analysis of a distributed topology graph pattern matching protocol, and iii) a systematic comparison of DASH video streaming implementations. Our experiences confirm that MACI provides support for the recurring tasks in the evaluation of diverse communication systems and significantly increases research efficiency. The experiments with MACI, i.e., the ProgMP, the topology graph pattern matching, and the DASH experiments, go beyond an evaluation of MACI and significantly contribute to the understanding of these domains. Overall, this dissertation contributes i) three programming models for the domains of Multipath TCP, adaptive communication systems, and topology adaptations in communication systems, ii) 13 novel, deployable general purpose, preference-, and application-aware Multipath TCP schedulers, and iii) a reusable framework for the seamless execution and analysis of extensive network experiments