Novel Decentralized Node Failure Recovery Scheme in Virtualized Networks

Abstract

The current Internet infrastructure, managed by multiple stakeholders, is unable to respond to the increasing new demands of services as it is hard to put the conflicted objectives of multiple Internet service providers together. Network virtualization has emerged as a new paradigm that overcomes the above challenge by partitioning the physical network to multiple isolated virtual networks used by different clients. Network resources failure is one of the issues that affect the network performance, which in turn impairs the service offered to the clients. This thesis addresses the problem of recovering virtual networks (VNs) affected by a substrate node failure. A novel heuristics-based algorithm that efficiently reallocates new resources to the affected VNs after a node failure is proposed. In this algorithm, a manager substrate node executes a set of recovery steps to migrate all the hosted virtual nodes in the failed substrate node in addition to the virtual paths traveling across it. The proposed approach is executed in a distributed manner without any coordination from the central infrastructure provider. The developed scheme efficiently minimizes the node failure recovery cost, the time needed to recover the virtual nodes hosted on the failed substrate node and hence significantly reduces the service interruption period. This, in turn, results in increasing the service provider revenue and decreasing the penalty charges paid for service level agreement violation. Performance results demonstrate the significant reduction in VN service cost and interruption time.