A model‐driven approach for deployment descriptor design in network function virtualization

• Published in: International journal of network management 2022-01, Vol.32 (1), p.1144-n/a
• Main Authors: Atoui, Wassim Sellil, Assy, Nour, Gaaloul, Walid, Ben Yahia, Imen Grida
• Format: Article
• Language: English
• Subjects: Computer Science; Configurations; Guidelines; Machine learning; Mathematical models; Virtual networks
• ISSN: 1055-7148; 1099-1190
• DOI: 10.1002/nem.2165

Summary Service providers in network function virtualization usually design manually or with static automation the deployment descriptors for virtual network functions. The descriptors are semi‐structured files that contain information about the resource requirements and the operational behavior of virtual network functions. Designing the descriptors manually and without formal strategies is certainly a cumbersome and error‐prone task for service providers. In this work, we propose a model‐driven approach that assists service providers in designing the deployment descriptors. This approach uses a configurable model iteratively to give service providers insights on which best configuration to choose. Concretely, we propose (1) to use a configurable deployment descriptor model, (2) a learning approach based on machine learning to automatically construct the configurable model, and (3) an approach that learns configuration guidelines from a catalog of deployment descriptors to assist service providers with the selection of the configuration to use. The configurable deployment descriptor model captures the relation and also the variability between the virtualized network function (VNF) elements from different deployment descriptors. We propose a learning approach to build the configurable deployment descriptor model by finding and federating similar VNF elements from different deployment descriptors. With our machine learning approach, we construct automatically the configurable model from a set of deployment descriptors. We use afterward the configurable model to learn configuration guidelines from the deployment descriptors and recommend them for service providers. The results of our experiments highlight the effectiveness of our approach to learning configurable deployment descriptor models.