Shared Protection-Based Virtual Network Embedding Over Elastic Optical Networks

This paper proposes a survivable virtual network embedding model over elastic optical networks considering shared protection against any single substrate node or link failure. A virtual network request is embedded in the substrate network with allocating the primary and backup resources which are node-disjoint. Modulation selection and spectrum allocation with constraints from elastic optical networks are considered in the embedding procedure. We consider the backup computing and transmission resource sharing to reduce the required backup resources. We formulate the proposed model as an integer linear programming problem to minimize the rejection ratio for a given set of virtual network requests. We introduce a greedy-based approach that promotes resource sharing to handle the larger-size problem in a practical time. In order to further improve the performance on the objective value, a deep reinforcement learning-based approach with polynomial time complexity in each episode is developed to solve the problem in multiple stages. We design a dedicated learning agent for each stage considering the problem property. We analyze the usages of different approaches based on performance evaluation. The numerical results show that, compared to a conventional model that adopts dedicated protection, the proposed model with shared protection reduces the rejection ratio by 15% on average in our examined cases.