Forecasting assisted VNF scaling in NFV-enabled networks

As a new promising technology, Network Function Virtualization (NFV) converts hardware based network function into software module running on virtual machines, for cost reduction and ease of management. These virtualized networks functions (VNFs) are commonly organized together as service function chains. Properly deploying VNFs is a key to achieve NFV targets. Most of existing efforts focus on one-time placement, ignoring the dynamic deployment and scaling needs of VNFs for the time-varying system. In this paper, we study the dynamic deployment and scaling of VNFs for operation cost minimization. We first formulate an offline VNF deployment cost minimization problem and prove its NP-hardness. Then, considering the dynamics of the network, we propose an efficient online scaling algorithm, which is composed of two parts: 1) One is Fourier-Series-based forecasting approach to minimize cost by avoiding frequent changes in network topology and 2) the other is online deployment algorithm to properly deploy VNF instances. We finally evaluate the proposed algorithms and results show that our algorithms can reduce more than 20% cost while maintaining the same system performance as other heuristic algorithms.