Traffic Engineering Based on Reinforcement Learning for Service Function Chaining With Delay Guarantee

Abstract

Network Function Virtualization (NFV) is an approach that provides a network service provider with agility and cost-efficiency in managing 6G network services. Standard traffic engineering rules are known limited in assuring a very stringent delay requirement in NFV when a traffic flow is required to follow a sequence of network functions scattered in data center networks. This paper proposes an innovative model and algorithm of traffic engineering for service function chaining (SFC) to maximize cost-efficiency under a delay-guarantee constraint. We first formulate the problem as a mixed-integer linear programming model for obtaining the optimal solution. We then propose an algorithm based on the reinforcement learning principles for finding an approximation solution in a large-scale problem with the dynamics of service demands. Numerical results under both real-world datasets and synthetic network topologies demonstrate that our proposed model and algorithm allow an NFV service provider (NSP) to place a virtual network function and steer a traffic flow efficiently in terms of energy cost for a delay-guarantee SFC. Importantly, the results provide an insight into the optimal and approximation solutions for an NSP to select a suitable traffic engineering approach with regard to network dynamics.