A four-stage adaptive scheduling scheme for service function chain in NFV

Abstract

Network Function Virtualization (NFV) enables the flexible software implementation of Network Functions (NFs) which is called Virtualized Network Function (VNF) and placed along the routing path of the network flow. A sequence of VNFs constitutes a Service Function Chain (SFC) to satisfy the processing requirements of flows. Since the SFC scheduling depends on the current network state and the dynamics of flows, it brings a great challenge to make an optimal SFC scheduling decision efficiently. In this paper, we present a Four-stage Adaptive Scheduling Scheme (FSASM) to make a trade-off between different scheduling goals and effects on network performance and management overhead. We design the specific mechanism for each stage when the network is in different workloads. Then, we prove the NP hardness of the optimization models in FSASM and propose a Minimum wEight Path Selection Algorithm (MEPS) with polynomial time complexity to realize a practical SFC scheduling. Moreover, we perform comprehensive experiments under different real-world topologies and network states. The results demonstrate that FSASM can achieve high network throughput and resource utilization as well as decrease the scaling frequency in highly dynamic network scenarios.