Resilient virtual network mapping against large-scale regional failures

摘要

Network Function Virtualization (NFV) and Software-Defined Networking (SDN) technologies enable a fast way of deploying network services by instant initialization of virtual network functions on standardized commodity hardware, and a flexible network control and management via dynamic reconfigurations over open protocols. In particular, virtual networks from distinct tenants can be provisioned on the same physical network infrastructure to share the computing and networking resources through a process called virtual network mapping, which can lead to an efficient utilization of the underlying physical resources while guaranteeing the service isolation and performance efficiency. In the process of provisioning virtual networks, it is critically important to ensure that the virtual network services are resilient and they have the capability to continue functioning well in case of various failures such as fiber cuts, natural disasters and malicious attacks. In this paper, we study the problem of resilient virtual network mapping against large-scale regional failures, which are more challenging than existing network survivability designs in terms of single node/link failures or shared risk link group failures. We propose a novel region-disjoint mapping (RDM) algorithm which can map the primary and backup virtual networks into nonoverlapping geographical areas to survive large-scale regional failures. More specifically, the RDM algorithm adopts (1) the backtracking techniques to ensure the disjointness of the primary and backup virtual networks, and (2) the Suurballe's algorithm to jointly optimize the mapping of primary and backup virtual networks in one-step. Simulation results show that the proposed RDM algorithm achieves much lower physical network resource consumption and blocking probability.