Optimization Model and Algorithm for Dynamic Service-Aware Traffic Steering in Network Functions Virtualization

Abstract

Network Functions Virtualization (NFV) has emerged as a paradigm for efficient, flexible and agile network function provisioning. In such NFV-based networks, ensuring network performance and cost efficiency is an important challenge to tackle when network traffic is steered through a chain of virtual network functions (VNF). In this work, we consider the dynamics of traffic demand in different time periods, and multipath routing for minimizing the routing cost in NFV. We focus primarily on optimization models and algorithms for finding a traffic steering solution that effectively splits a demand volume into several flows and selects appropriate links and nodes for these flows. We formulate the problem as a mixed linear integer programming model for obtaining an optimal solution taking into account the dynamics of service demand, multipath routing and service function chaining. For the large scale problem, we propose a heuristic algorithm to find an approximation solution. Particularly, our proposed model and algorithm allows a controller to update a link weight system for effectively steering traffic demand to appropriate nodes in a NFV infrastructure. The evaluation results show that our approach to traffic steering significantly improves a number of major performance metrics including the routing cost, the maximum link utilization, and the accepted demands. In addition, the approximation solution is very close to the optimal solution.