Reliable embedding of virtual networks in Cloud's backbone network

Overview & Synopsis

Cloud computing has known an impressive success over the last few years and is continuously emerging personal and profesional life thanks to its elasticity, pricing model, etc. This innovative technology has attracted both industrial and research communities and becomes omnipresent.In order to take benefit from the Cloud expansion, providers require effecient management strategies to properly supply their physical capabilities such as network resources. Besides, Cloud providers have to respect the Service Level Agreements (SLA) and avoid any outage impact in order to guarantee the Cloud-based service continuity. In this thesis we tackle the problem of reliable virtual network embedding within the Cloud backbone by considering the impact of physical equipments' outages. Our main focus is to improve the provider's turnover by i) maximizing the acceptance rate of the incoming virtual networks issued from clients' requestand ii) minimizing the penalties induced by service disruption due to the physical failures. This optimization problem is NP-hard with multi-objective and non-linear formalization. To cope with this complexity and since reaching the optimal solution is computationally intractable, we propose different strategies that aim to respect the aformentionned objectives. First, we propose a preventive approach named PR-VNE that urges the use of reliable network resources in order to avoid the physical failures impact.PR-VNE strongly relies on the Artificial Bee Colony metaheuristic to reach an optimized solution. It should be highlighted that PR-VNE does not adopt a recovering mechanism to deal with the network outages. Second, we devise a new reactive approach named CG-VNE that does not consider any backup resources but re-embed the impacted virtual resources once an outage occurs in the underlying network. As well as this reactive mechanism, CG-VNE adopts the same preventive strategy like PR-VNE by avoiding the unreliable resources. It should be noted that CG-VNE is devised basing on a Game Theory framework by defining a collaborative mapping game. Finally, we deal with the survivable batch mapping problem that considers the embedding of a virtual networks set instead of one client request. We introduce a new reliable batch embedding strategy named BR-VNE that relies on Monte-Carlo Tree Search algorithm.BR-VNE delegates the embedding of one virtual network request to any online algorithm and focuses to find the best mapping sequence order. The performance evaluation of our algorithms leads to efficient results.