Experimental Evaluation of Orchestration Software for Virtual Network Functions

The adoption of Network Functions Virtualization (NFV) is considered as one of the enablers for a fully softwarized 5G architecture, that allows significantly higher flexibility for network service providers to instantiate services, conFigure and update them, as well as to truly realize multi-tenancy. The key enablers of the NFV technology lie in the evolution of virtualization technologies, such as the long established Virtual Machines, and the more recent paradigms of containers for micro-services, like docker and LXC. Such technologies allow the virtualization to span even to the wireless RAN, with fully softwarized base station architectures. Nevertheless, with the plethora of different virtualization technologies, several Virtual Infrastructure Managers (VIMs) and service orchestrators have emerged, each of them addressing different aspects for the provided services e.g. possible nomadic behaviour of the hosting computers, resource constrained devices hosting the services and services deployed for time critical services to name a few. In this work, we use a reference setup and experiment with some of the most widely adopted infrastructure managers, trying to experimentally derive and validate their differences under varying loads of traffic. Our results reveal the time needed for instantiating the same service function, for dockers, containers and Virtual Machines, for a different number of VNFs.