EAGER: Towards A New Microservices-based Architecture for Network Functions Virtualization

点击翻译按钮获取中文摘要

Network functions virtualization (NFV) promises to allow network operators to deploy new network functions (NFs) rapidly, quickly respond to external events, and ensure high performance and robustness. Thus, NFV can offer greater agility, scalability and performance-optimality than legacy networking gear. Many entities---including data centers, Internet Service Providers, telcos, and cellular providers---are espousing NFV. A key aspect that determines whether the promise of NFV can be fulfilled is the architecture used to realize NFV and the performance of its implementation. Current NFV architectures are closely following the trends corresponding to Web services where a popular approach is to use a microservices-based architecture: in this, network functions are deployed within light-weight containers running atop barebones operating systems, and managed using cluster schedulers. These microservices-based architecture is purported to offer agility and scale-out performance. While microservices are well suited for Web services that are largely request-response based, they are a poor fit for NFV because of the impact on both control and data planes. High latency is a natural impediment in such microservices-based architectures. But, a variety of other problems pertaining to workload agility, availability and throughput, and isolation can also arise. In this project, the PI will conduct an in-depth empirical study of the trade-offs of existing microservices-based architectures. Based on these results, the PI will conduct research into two alternate architectures, a strawman based on stream processing, and a clean-slate approach based on an extensible exokernel-like design. Finally, the PI will develop mechanisms to specify network functions processing and runtime composition, NF addition, and deletion. This project's intellectual merit lies in the fact that it will lead to key insights into the differences and commonalities between Web services and NFV when it comes to realizing them using microservices. Furthermore, it will lead to insights on how to create ground-up microservices-based frameworks that are optimized for NFV. The outcomes of this initial exploration can inform future research into the fundamentals of NFV including security and isolation in emerging NFV architectures, NFV programming languages, and verification. Broader Impact: The project will also have significant broader impact. The PI plans to engage a postdoctoral researcher in this research. Thus, the research will give rise to several professional development opportunities for the postdoctoral researcher involved. Outcomes from this research will be made available as profiles on CloudLab (a public cloud testbed) for other researchers to leverage. The research will also be tightly integrated into courses at the University of Wisconsin-Madison.