Abstract :
To support real-time and security-demanding applications (e.g. telepresence, cloud computing) at a large-scale, the Internet must evolve so that Network Service Providers (NSPs) provide end-to-end Quality of Service (QoS) across their networks. The delivery of such QoS-assured services requires the negotiation of end-to-end QoS contracts (Service Level Agreements, SLAs) among NSPs and the configuration of their networks accordingly. The management of inter-NSP SLA negotiation is usually treated as an optimization problem, assuming that NSPs cooperate and agree on a common system, providing a solution for each demand. This assumption is quite strong in a highly competitive context where NSPs are cautious about sensitive data disclosure like topology or resource usage information or even SLA descriptions and prices. Hence, to meet NSPs´ requirements on confidentiality, we opt for a distributed framework. In order to not over-provision demands, we consider the problem in a wider range: not only on the basis of instantaneous requests but also anticipating future ones. To enhance the chance of an NSP to be selected for an end-to-end service, we aim to take into account the demander likeliness of acceptance (aka. customer utility). To this end, we opt for reinforcement learning techniques and propose three distributed algorithms, inspired by the Q-learning algorithm, having different cooperation levels.
Keywords :
Internet; computer network security; contracts; data privacy; distributed algorithms; learning (artificial intelligence); optimisation; quality of service; Internet; Q-learning algorithm; confidentiality requirement; cooperation level; distributed algorithm; distributed framework; distributed learning algorithm; end-to-end QoS contract; end-to-end quality of service; inter-NSP SLA negotiation management; network service provider; optimization problem; reinforcement learning; security-demanding application; service level agreement; Internet; Learning systems; Optimization; Quality of service; Real-time systems; Service level agreements; Reinforcement learning; SLA negotiation; distributed algorithm; inter-NSP;
