Optimization of Migration Cost for Network Function Virtualization Replacement

Abstract

Today's networks are concerned about making the control of communication flexible and improving the existing management systems in such a manner that reduces the Capital expenditures (CAPEX) and operating expenses (OPEX), through reducing equipment costs and energy efficiency. Along with the benefits of decreasing the time to promote new services to the clients, service providers' attention has gradually moved to Network Function Virtualization (NFV), which is a potential technology decoupling network functionalities from hardware and is a promise of high performance service provision with optimizing resource utilization across various infrastructures. However, to simultaneously achieve these goals, sometimes it is necessary to instantiate a new function depending on the traffic pattern of high-bandwidth characteristics and Quality of Service (QoS) measures. Due to the limited resources at the node, other functions in the node may need to be migrated to other nodes in order to provide resources for the new functions. Existing works related to the Virtual Network Function (VNF) deployment and migration usually focus on proposing new deployment strategies and migration mechanisms. However, reducing migration cost restricted to memory, CPU, and bandwidth capacities is not considered in those studies. In this work, the problem of virtual network functions migration is formulated as an Integer Linear Program (ILP) with the objective of minimizing the migration cost while satisfying computing and network resource capacities constraints and selecting the minimum cost path from the source to the destination node. Since the ILP is NP-complete, we propose a greedy minimum migration cost (GMMC) algorithm. Simulation results show that the proposed GMMC algorithm can reduce the total migration cost by up to 61% and the number of migrations by up to 52% when compared to the state-of-the-art schemes.