ClusVNFI: A Hierarchical Clustering-Based Approach for Solving VNFI Dilemma in NFV Orchestration

Network functions virtualization (NFV) is becoming a prevailing design for future Internet by migrating network functions from dedicated hardware appliances to software instances running in virtual computing platforms. NFV resource allocation approaches can dynamically instantiate network functions by using virtualized network functions (VNFs) to satisfy various quality-of-service (QoS) requirements with minimum network costs. The operator can launch a new VNF instance (VNFI) for each VNF required for flows, or assign it to the established VNFI. This makes NFV orchestration (NFVO) even more complicated. In addition, the challenges of developing NFVO scheme include how to manage the dependency between VNFs placement (network-level) and routing of flows (flow-level) through ordered VNFs, and how to efficiently utilize the available network resources. In this paper, ClusVNFI, a hierarchical resource allocation approach based on clustering, is proposed to address these challenges. To be specific, VNFs are proposed to be clustered based on their correlation. Then VNFs belonging to the same cluster are inclined to be deployed on one node to reduce the occupied link bandwidth. Moreover, network nodes are clustered based on the similarity on end-to-end flow latency information. Accordingly, flows in the same cluster intend to share the instantiated VNFIs, aiming at improving VNFI utilization while avoiding path stretch. By capturing the dependency between network-level and flow-level through clustering, ClusVNFI can achieve the tradeoff among multiple objectives including maximizing the number of admitted flows, minimizing path stretch, and improving VNFI utilization. Extensive simulation results show that the proposed ClusVNFI can balance multiple objectives comparing with other typical heuristic algorithms. Moreover, ClusVNFI can reduce network resource occupation effectively, while guaranteeing the average delay and network hops.