Probabilistic inference-based service level objective-sensitive virtual network reconfiguration

Abstract

Network virtualization enables multiple service providers to share the same physical infrastructure, and allows physical substrate network (SN) resources to be used in the form of a virtual network (VN). However, there are many obstacles to the application of this technology. One of the more challenging is the reconfiguration of SN-embedded VNs to adapt to varying demands. To address this problem, we propose a service level objective (SLO)-sensitive VN reconfiguration (VNR) method. A Bayesian network learning and probabilistic reasoning-based approach is proposed to automatically localise reconfiguration points and generate VN resource requests. To determine an optimal reconfiguration solution, we design a heuristic VNR algorithm with a virtual node and virtual link swapping strategy. We validate and evaluate this algorithm by conducting experiments in a high-fidelity emulation environment. Our results show that the proposed approach can effectively reconfigure a VN to adapt to a changed SLO. A comparison shows that our reconfiguration algorithm outperforms existing solutions.