A novel evaluation function for higher acceptance rates and more profitable metaheuristic-based online virtual network embedding

• Published in: Computer networks (Amsterdam, Netherlands : 1999) Vol. 195; p. 108191
• Main Authors: Aguilar-Fuster, Christian, Rubio-Loyola, Javier
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 04-08-2021; Elsevier Sequoia S.A
• Subjects: Acceptance; Algorithms; Combinatorial analysis; Cost analysis; Embedding; Evaluation functions; Fitness landscape; Heuristic methods; Network topologies; Optimization; Revenue; Search process; State-of-the-art reviews; Substrates; Virtual network embedding; Virtual networks; Evaluation functions; Fitness landscape; Virtual network embedding
• ISSN: 1389-1286; 1872-7069
• DOI: 10.1016/j.comnet.2021.108191

Virtual network embedding (VNE) is a challenging combinatorial optimization problem that has been widely addressed through metaheuristics. However, one key element for the successful application of metaheuristics in the VNE problem is the evaluation function, which should be pivotal to making fine distinctions among all the potential embeddings to avoid producing zones with solutions of the same fitness value (i.e., plateaus zones). The existence of plateaus zones in the search space is a critical aspect that would jeopardize the detection of promising search directions in metaheuristic-based VNE. In this paper, we demonstrate that the classical VNE evaluation function based on the embedding cost has poor discrimination capabilities, namely it fails in establishing preferences among the potential embedding solutions. To cope with this important issue, we propose the Alternative Function Based on Degree (AFBD), which takes into account the degree of the nodes of the substrate network to drive the search process. Our results demonstrate that VNE based on the AFBD objective function outperforms the state-of-the-art evaluation functions in terms of the acceptance rate and revenue-to-cost metric. Using AFBD as a reference, we report performance decreases in VNE algorithms with the state-of-the-art evaluation functions ranging from 1% to 7% (relative) for the acceptance rate and decreases close to 12% for the revenue-to-cost metric, which means that the AFBD-based VNE algorithm accepts more virtual networks at lower embedding cost. Formal analysis demonstrates that it is possible to enhance the performance of metaheuristic-based VNE through the use of the AFBD objective function, even with substrate network topologies where the performance of metaheuristic based-VNE decreases drastically.