Automated generation of VNF deployment rules using infrastructure affinity characterization

As the rollout of NFV in the Telecommunications domain gains momentum, methods that enable the automated and performant deployment of VNFs are required to support large scale deployments. This paper presents a framework that enables the automated characterization of VNFs and modelling of the relationships between resource types, quantities and achieved performance. A workload characterization approach is proposed to identify the most appropriate types of resources to be allocated to a VNF at deployment time. Machine learning modelling was used to identify the appropriate deployment configuration which meets required performance targets, such as network throughput. The proposed framework has been experimentally validated based on the optimization of a Traffic Classifier VNF deployment in an OpenStack cloud environment.