No abstract

No abstract

The communication networks of today can greatly benefit from autonomous operation and adaptation, not only due to the implicit cost savings, but also because autonomy will enable functionalities that are infeasible today. Across industry, academia and standardisation bodies there has been an increased interest in achieving the autonomous goal, but a path on how to attain this goal is still unclear. In this paper we present our vision for the future of autonomous networking. We introduce the concepts and technological means to achieve autonomy and propose an architecture which emerges directly through the application of these concepts, highlighting opportunities and challenges for standardisation. We argue that only a holistic architecture based on hierarchies of hybrid learning, functional composition, and online experimental evaluation is expressive and capable enough to realise true autonomy within communication networks.

The stability and the predictability of a computer network algorithm's performance are as important as themain functional purpose of networking software. However, asserting or deriving such properties from thefinite state machine implementations of protocols is hard and, except for singular cases like TCP, is notdone today. In this paper, we propose to design and study run-time environments for networking protocolswhich inherently enforce desirable, predictable global dynamics. To this end we merge two complementarydesign approaches: (i) A design-time and bottom up approach that enables us to engineer algorithms basedon an analyzable (reaction) flow model. (ii) A run-time and top-down approach based on an autonomousstack composition framework, which switches among implementation alternatives to find optimal operationconfigurations. We demonstrate the feasibility of our self-optimizing system in both simulations and real-world Internet setups

No abstract