VNFAccel: An FPGA-based Platform for Modular VNF Components Acceleration

The expected benefits of Network Function Virtualization (NFV), i.e., flexibility and efficiency, require platforms able to instantiate Virtualized Network Functions (VNFs), replacing traditional dedicated middleboxes. However, to fulfill performance requirements, especially for functions operating at high-throughput line rates, dedicated hardware acceleration may be required and, to provide the expected flexibility, hardware accelerators must be able to support different VNFs. Thus, reconfigurable devices such as Field-Programmable Gate Arrays (FPGAs) have received attention due to their ability to provide both properties. Not all VNFs, however, require or are even suitable for hardware acceleration and, therefore, heterogeneous platforms composed of FPGAs and General Purpose Processors (GPPs) can be used. Even for a single network function, distinct VNF Components (VNFCs) may have different demands and restrictions, meaning that different hardware substrates may be more suitable for their implementation. In this paper, with the goal of promoting the efficient and seamless integration of accelerators in NFV, we propose VNFAccel, a platform to manage the execution of VNFCs in a heterogeneous infrastructure. We evaluate the effectiveness of FPGAs as accelerator devices for two VNF case studies and analyze the instantiation latency of FPGA-based VNFCs in the platform. We also provide guidelines to design VNFCs that are more flexible, promoting increased reuse and reducing configuration time, and identify properties that make them suitable for FPGA implementation.