Closing the gap between performance and efficiency in programmable networking

In the last decade, SDN and NFV paradigms have been proposed to alleviate the problem of the ossification of the Internet architecture, but despite the flexibility achieved, both scenarios are showing deficiencies in jointly achieving performance and resource efficiency. And as the complexity of network functions continues to increase, this technical limitation is arising nowadays. In this scenario, this Dissertation aims to bind versatility, computing capabilities and efficiency in the implementation of network functions through programmable networking technologies. In particular, this work proposes two strategies able to exploit new challenges and requirements brought out by a new hardware-driven research trend based on programmable networking abstractions. Firstly, we focus on networking platforms by proposing a programming stateful abstraction, called FlowBlaze, able to program the hardware level for the execution of stateful network functionalities at high speed without requiring no hardware design expertise. Furthermore, we challenge the proposed abstraction through design and deployment of typical network-related tasks, such as ensuring session continuity in railways, as well as functions for the computation of upper layer functions (e.g. data aggregation). Secondly, we exploit data structures and algorithms to design and propose ad hoc solutions. Indeed, ad hoc solutions are designed to be tailored to the characteristics of the network task to perform, allowing to optimize and reduce the gap between performance and efficiency. Inside this strategy, we focus on two important network tasks such as packet classification and network monitoring by per-flow distinct counting. In the former case, we propose a new online scheme, called TupleMerge, able to reduce classification time while preserving similar performance in rule updates. In the latter case, we propose a high performance - low memory top-k spreader detection algorithm called FlowFight, other than several optimizations to enhance state-of-the-art sketches.