Network Functions Virtualization (NFV): Challenges and Deployment Update

Enter The Software-Defined Era

The integration of Information Technologies and Communications, commonly referred as ICT, has been more a long-term goal than a reality for a long time. Roughly speaking, networking and computing knew an evolution at comparable pace till the global availability of the Internet and the almost pervasive application of its protocols and the architecture supported by them to any networking problem. Not surprisingly, the very success of the Internet basic technologies made the evolution of the technologies applied for providing end-to-end connectivity more and more difficult, precisely because these network technologies were the base for the radical changes that were taking place in the IT arena, around the ideas of Internet-based services and, most of all, the cloud.

IT evolution and, in particular, the evolution of Internet-based services, have been rooted on successive revolutions in software development practices, and in more and more powerful abstractions easing their conception, creation, and operation. This software-based nature has allowed business actors in the IT services arena to become agile in terms of satisfying new requirements and deploying new solutions, best exemplified by the DevOps approach in Loukides (2012), a set of best practices gaining strong momentum in the IT industry and focused on the tight communication (and even integration) of the activities related to development, operations, and quality assurance.

Network infrastructures, on the other side, became tied to their topologies and the requirements on using open, standard interfaces among the different nodes in these topologies. While the development of network nodes became certainly software intensive, the evolution of network services was tied to longer innovation cycles, requiring the agreement on standards among node vendors for interoperability purposes. What is more, the generalization of the network node as the basic functional unit implied an enormous degree of heterogeneity in network elements and their management procedures, which translated into additional problems for any attempt to make network infrastructure evolution agile or able to easily satisfy evolving user requirements.

On the Internet arena we had on the one hand network service providers, dealing with highly-heterogeneous and difficult to evolve infrastructures, and unable to address in a timely manner specific user requirements or to cover long-tail demand at a reasonable cost, while on the other hand there were the IT service providers, much more agile, relying on an almost uniform infrastructure, and able to adapt and evolve their software at a much faster pace, therefore being able to increase their value while network infrastructures were becoming increasingly ossified.

The advent of cloud computing was not only a forward step in IT service virtualization, as they did not need to be hosted at a physical infrastructure operated by the IT service provider anymore, but at the same time it implied some additional requirements on the network infrastructure that could not be solved by legacy (networking) techniques so far. These additional requirements on flexibility (i.e., the network had to adapt to the ever-changing cloud configuration) and abstraction (as applications needed to interact with the network as a resource among others) brought the need for a different conceptual framework for networks, beyond the usual approaches based on physical devices hosting a fixed, limited set of functionalities. Furthermore, the idea of running IT services on virtualized infrastructures made some researchers think about the possibility of doing the same for the functions performed in the network nodes, resulting in what will be referred to in the following as network functions. These two orthogonal directions constituted the basis for the current trends in network softwarisation.

Network softwarisation is a general term referring to all techniques oriented towards the application of two main and related principles:

• •

  Providing a general interface for the provisioning, management, control and invocation of network resources, by means of software abstractions that hide complexity and deployment details of actual network infrastructures.

• •

  Decoupling the different planes composing the network (data, control and management), and using open interfaces between them, in order to make the supporting infrastructure as much regular and homogeneous as possible and relying on software mechanisms to support specialized functionalities.