Optimisation de routage et d'allocation de ressources basée sur le SDN dans les réseaux de data centers pour les architectures de cloud privé
Roua Touihri · HAL (Le Centre pour la Communication Scientifique Directe) · 2021
According to the latest statistics, the number of connected people to the Internet is still exponentially growing and the high quality of cloud services will remain significantly requested in the coming years. Facing the resulting tremendous growth of the intra-data center traffic, the traditional Data Center Network (DCN) architectures are not capable to stay ahead of the demand in terms of scalability and cost lowering. Besides, DCNs suffer from the degradation of their network Quality of Service (QoS) performances which deeply impact the users Quality of Experience (QoE). As a result, cloud architects and service providers are pressed to deal with this rapid explosion of data center traffic by reconsidering their data center architecture to fit their customers’ needs. Hence, implementing new techniques based on optimized algorithms become compulsory to overcome these challenges within data center networks. In this regard, we tackle the performance of data center networks by investigating new promising approaches able to improve intra-data center communication and to optimize the resource allocation within their infrastructures. Motivated by its architecture strengths, we study in this thesis the CamCube Server-Only data center managed by ONOS SDN controller. Besides, we aim to implement and analyze the performance of strategies to address the optimization of routing and resource allocation for intra-CamCube servers communications. In fact, our objective consists of improving network performance and decreasing network congestion. The problem of resource allocation and routing in SDN-based CamCube topology is NP-hard. In order to overcome this challenge, we investigate the problem through three increasing complexity steps. In the first stage, we consider unicast intra-CamCube communication by proposing new methods named CRP and ACOCRP respectively based on Linear Programming Ant Colony Optimization. The latters generate an optimized path to forward packet in a typical CamCube network subjects to the constraints of network performance in terms of latency and path length. In the second phase, we propound a new M-CRP and ACO-MCRP approaches to tackle the multicast communications in SDN-based CamCube network. Our approaches addressed multicast routing and resource allocation of online arrived flows. We show that our propounded approach enhanced the CamCube QoS and the quality of the proposed multicast tree. In this thesis, we online treat the routing and the resource allocation for the communication of each arrived flow. As a third step, we focused on resource allocation optimization for batch mode arrival of traffic flows within ONOS-based CamCube topology by proposing batch-(M)CRP strategy. Note that we emulated the propounded environment to test the full proposed schemes, through extensive experimentations conducted with Mininet. We evaluate the performance of our propositions in terms of E2E delay, jitter and packet loss. Obtained results demonstrate that our proposals outperform the existing state-ofthe-art strategies such as the shortest path and OSPF routing schemes respectively for CamCube and traditional Clos DCNs architectures.