AbstractThe increasing number of Internet users around the world is putting strain on the rigid architecture and the limited flexibility of traditional networking systems. This has led to the need for new highly adaptable, programmable and easily manageable networks. In this regard, the new architecture paradigm of Software-Defined Networking (SDN) has emerged over the past few years with an aim to simplify the network management and overcome the inflexibility of the traditional networks. SDN has become a hot and thriving topic due to its advantages, which have attracted the attention of both the industrial and academic sectors. However, a review of the relevant literature outlines some concerns regarding the dependability and fault management of SDN.
Given that SDN is centralised networking system, this means that the network controller is responsible for managing all the network activities. Network link failures occur in everyday operation and it represents 70% of all IP backbone failures. Every time a link goes down, the SDN controller will need to update the network. The current SDN restoration mechanisms do not take into account the time required to update and the cost associated with it, which is important to reduce the recovery process. Moreover, none of the previous work has considered the service disruption, which is highly expected after every failure scenario and
leads to decrease the network service availability.
This work focuses on two different aspects. These are accelerating the restoration
of failure recovery and improving the network service availability. The contributions of this thesis are twofold. First, two new approaches are proposed to provide a significant improvement on existing SDN restoration scheme while maintaining reasonable cost. The approaches rely on the community detection and path anatomy concepts. For both approaches, new network models, algorithms and frameworks are designed. Second, unlike the existing
literature, we propose new method that utilises the failure forecasting technique to enrich and facilitate fault management mechanisms of SDNs. This technique allows the SDN controller to predict and avoid risky paths by finding alternative ones before the failure incidents occur. Therefore, this approach aims to decrease the service disruption, which in turn increases network service availability. For validation of the technique; a new network model, a failure generation model, a risk analysis, two algorithms and a framework are designed.
All the proposed methods have been implemented using popular simulation tools and experimental results are presented to show how the proposed methods and algorithms can work effectively to enhance the performance of SDN fault management, in particular, and its dependability, in general. Finally, in light of the findings of this work, suggestions are made towards achieving more advances in this research area.
|Date of Award||8 Jul 2019|
|Supervisor||Benjamin Aziz (Supervisor), David Bryan Carpenter (Supervisor) & Mo Adda (Supervisor)|