CAREER: A Runtime Framework for Network Function Virtualization

University Of California-Berkeley, Berkeley CA

Investigators

Abstract

The Internet has transformed everyday life across the globe. However, while there has been constant evolution in the applications that use the Internet, the same is not true of the underlying Internet itself. Instead, the Internet's core service offering has proven tremendously difficult to change and, in fact, has remained largely unchanged since the 1980s. This is problematic because of its impact on the network 'carrier' companies that own and operate the network infrastructure: any upgrade or change in their networks is a tremendously difficult and expensive process, leading to partial and ad-hoc fixes and escalating management costs. The difficulty of effecting change also means that network carriers have been stifled in their ability to explore new services and business models. In short: the long term health of the Internet depends on the health of the carriers that operate its underlying infrastructure, but these carriers are increasingly squeezed between mounting costs and flat revenues. This trajectory is now widely recognized as unsustainable. The research being undertaken in this project offers a way out of this bind by making efficient software traffic processing a fundamental component of the network infrastructure. Traditionally, network infrastructure has been built from specialized fixed-function hardware which is slow and expensive to evolve. Augmenting this infrastructure with software functions that run on commodity general-purpose hardware, will allow carriers to more easily and cheaply evolve their existing services, as well as to explore innovative new service models. In other words, this project's research will allow network functionality to evolve on (faster) software rather than (slower) hardware development timescales. Today, the de-facto approach to incorporating advanced traffic-processing functionality into a network is through the deployment of specialized 'middleboxes'. These middleboxes have grown to be a vital part of modern networks but are also widely recognized as bringing with them significant problems including high cost, inflexibility, and complex management. Recently, advances in software packet processing and virtualization technologies has led industry and research to explore an alternate approach: moving middle- box functionality out of dedicated boxes into software applications that run multiplexed on commodity x86 processors. This approach - termed Network Function Virtualization (NFV) - stands to reshape network infrastructure and bring the benefits of cloud computing to networks. However, the evolution of cloud computing (particularly for data analytics) has greatly benefited from the existence of frameworks that implement application-independent methods for elastic scaling, placement and failover, achieving high efficiency while relieving programmers of these burdens. NFV has no such general management solutions. The central premise of the research agenda being undertaken in this project is that NFV needs a framework, by which we mean a software environment for packet-processing applications that implements general techniques for common functions. The project addresses fundamental questions that arise when processing network traffic in software. It provides general and effective solutions to the common problems of scaling, performance management, and fault-tolerance for software-based network functions. Moreover, by showing how these solutions can be embedded in a common framework, the research enables faster innovation: third-party vendors can more easily develop new application-relevant functions since they no longer have to deal with the common but thorny issues of availability/scalability/efficiency which are now handled by the framework. Broader Impact: The project contributes to the broader goal of ensuring that the Internet continues to serve as the vehicle bringing transformative communication services to the global population. The project's educational component addresses the growing divide between coursework and practice in networked systems through a new undergraduate course that uses the above NFV framework to expose students to modern network software and provide in-depth experience in building and deploying innovative network services. The project is also exploring lab methodologies to scale such project-centric classes to 100+ students. Outreach activities include the creation of an open-source effort in collaboration with network vendors and carriers, and mentoring undergraduates and under-represented students. For this, the PI builds on a strong record of working with industry to impact practice and an exceptional record of building a diverse research group.