Network Function Virtualization (NFV) is a promising technology that enables the transition of network functions (e.g. routers, CMTS, firewalls etc) from dedicated hardware appliances to general purpose servers. The first wave of NFV was nicknamed "lift-and-shift" because it was about taking a network function and running it as-is in a virtual environment. However, the Cloud-Native architectures that the web giants are using take a different approach to virtualization. They view the cloud as a huge distributed compute platform where "functions" are broken into micro-services and scheduled in a container based environment onto the available compute/storage resources.
The web-native approach view networking applications as just another type of software application and this is indeed the correct view of networking in this framework. But the devil is in the details and out-of the-box schedulers such as Kubernetes do not address the specific requirements of networking application, specifically when it comes to processing the data plane path.
This paper will highlight the differences between the old and new virtualization approaches, identify the gaps where the cloud native approach needs to be extended for networking and offer solutions. More specifically the presentation will compare and contrast the direction the industry takes with network function virtualization vs. cloud native.
This paper is high-level and outlines the main issues and comments of possible solutions without going into specifics.