Modelos e algoritmos para o problema de mapeamento de redes virtuais

Abstract

Network Virtualization has been used to allow that many virtual networks can coexist over a substrate network, sharing its resources. In this scenario, the problem of mapping the virtual networks onto the physical substrate, meeting capacity and quality requirements, emerges. In this work, a compact model based on integer linear programming, which provides optimal solutions, is presented. Two extended models are proposed, with stronger dual bounds than the compact model. Based on these models, we also propose branch-and-price algorithms. Results show that, for instances where geographical location is considered for the nodes, both algorithms achieved, within the time limit, better solutions than the compact model for a greater number of virtual network requests. Moreover, it is performed a study on the impact of different objective functions (mapping cost, delay, and load balancing) and quality of service constraints (geographic location, maximum delay, and resilience) on the solutions