A Framework Integrating FPGAs in VNF Networks

Abstract

Future telecommunications networks require more throughput and lower latency. The slowdown of Moore’s Law forces one to think beyond traditional CPU servers to satisfy these needs. FPGAs have been shown to provide exceptional throughput and latency, but they currently lack support for many of the orchestration and reconfiguration tools required by telecommunications networks, namely exposing a standard configuration API for SDN applications to control, and having the ability for tools such as Kubernetes to provision and reuse these compute resources.We propose an FPGA Framework for Interactive VNF Environments, or FFIVE for short. We utilize the framework for the creation of FPGA-based containers allowing them to be deployed by Kubernetes and virtually connected via VXLAN. This enables us to migrate the FPGA as needed in the virtual network, allowing for quick image substitution as usage changes. We also enable the use of RESTful APIs to control, reprogram, and manage an FPGA making it usable by SDN applications. Essentially, our framework makes the deployment and management of FPGA-based containers in telecom analogous to their CPU counterparts but with the improved bandwidth, efficiency, and latency provided by FPGAs. Comparing the network stack and chaining of our approach with conventional multi-threaded CPU server implementations, we offer 59-fold to 175-fold throughput increase and a 97.5% reduction of latency for networking and chaining. We are also able to reduce the latency of a real-world firewall VNF by 92.3%.