Offloading in P4 Switch Integrated with Multiple Virtual Network Function Servers

Abstract

Software Defined Networking (SDN) and Network Function Virtualization (NFV) are two transformative technologies that offer distinct benefits. SDN virtualizes the control plane by separating it from the data plane, while NFV virtualizes the data plane by moving network functions from hardware and implementing them in software. Therefore, combining SDN and NFV can fully exploit the benefits of both technologies. As Programming Protocol-independent Packet Processors (P4) become popular due to its flexibility, traditional SDN switches are being replaced by P4 switches. In the P4+NFV architecture, network functions can be provided in both P4 switches (PNF) and NFV servers (VNF). However, to minimize packet delay, the offloading problem between P4 switches and NFV needs to be addressed. The novelty of our paper lies in investigating the offloading problem and evaluating the impact of employing multiple VNFs with varying computing capacities within the P4+NFV architecture. We also use M/M/1 queuing theory to derive the average packet delay and propose an optimization solution based on gradient descent to find out the optimal offloading probabilities of various VNF servers. Results show that optimal offloading from P4 switch to multiple VNFs can reduce the average packet delay from 4.76% to 40.02%.

Keywords