MSV: An Algorithm for Coordinated Resource Allocation in Network Function Virtualization

The proposition of network function virtualization (NFV) aims to solve the difficulty and ossification in current network’s management and service provision caused by ever-growing NFs with dedicated hardware. By decoupling the NFs from dedicated hardware to virtualized platform, NFV promises flexible deployment and management of service function chains (SFCs). However, an optimal resource allocation for requested SFC in NFV-based infrastructures should coordinately consider following three stages: virtual network functions (VNFs) chain composing, VNF forwarding graph embedding, and VNFs scheduling, which is a tough task as the decision of these three phases is mutually dependent. In this paper, staring from the challenges in solving coordinated NFV resource allocation (NFV-RA), we first formulate a typical three-stage coordinated NFV-RA model as a mixed integer programming (MIP) and, then, propose a heuristic solution called merge–split viterbi (MSV). MSV can automatically determine the appropriate number of VNF instances without given maximum number threshold, and it does not take the iterative deployment strategy, which is commonly used in current solutions. The main idea of MSV is to first find a global basic solution and, then, to further optimize the basic solution through some improvement procedures, and this makes it not be easily trapped in local optimality and avoid complex anti-local-optimal measures as well. Extensive experiments demonstrate that MSV can get solutions in global range with reasonable execution time and achieves total cost ratio within 115% compared to the MIP implement.