Learning a Configurable Deployment Descriptors Model in NFV

Abstract

The deployment descriptors in Network Function Virtualization (NFV) are usually designed and configured through static automation and manual edition by service providers without any formal strategy except best practices. Thus, leading to an error prone and time consuming approach. We propose in this paper 1) a configurable deployment descriptor model and 2) a learning approach based on machine learning to construct the configurable model automatically. Firstly, the configurable deployment descriptor model captures the relation and also the variability between the VNF elements of different deployment descriptors. It enables service providers to configure and generate customized deployment descriptors instead of designing them each time from scratch. Secondly, we define a learning approach to learn configurable deployment descriptor models by finding and federating similar VNF elements of different deployment descriptors. With our machine learning approach we construct automatically a configurable model from a set of deployment descriptors. The results of our experiments highlight the effectiveness of our approach into learning configurable deployment descriptor models.