New information model that allows logical distribution of the control plane for software-defined networking : the distributed active information model (DAIM) can enable an effective distributed control plane for SDN with OpenFlow as the standard protocol

Publication Type:
Issue Date:
Full metadata record
Files in This Item:
Filename Description Size
01front.pdf286.44 kB
Adobe PDF
02whole.pdf8.09 MB
Adobe PDF
In recent years, technological innovations in communication networks, computing applications and information modelling have been increasing significantly in complexity and functionality driven by the needs of the modern world. As large-scale networks are becoming more complex and difficult to manage, traditional network management paradigms struggle to cope with traffic bottlenecks of the traditional switch and routing based networking deployments. Recently, there has been a growing movement led by both industry and academia aiming to develop mechanisms to reach a management paradigm that separates the control plane from the data plane. A new emerging network management paradigm called Software-Defined Networking (SDN) is an attempt to overcome the bottlenecks of traditional data networks. SDN offers a great potential to ease network management, and the OpenFlow protocol in particularly often referred to a radical new idea in networking. SDN adopts the concept of programmable networks which separate the control decisions from forwarding hardware and thus enabling the creation of a standardised programming interface. Flow computation is managed by a centralised controller with the switches only performing simple forwarding functions. This allows researchers to implement their protocols and algorithms to control data packets without impacting on the production network. Therefore, the emerging OpenFlow technology provides more flexible control of networks infrastructure, are cost effective, open and programmable components of network architecture. SDN is very efficient at moving the computational load away from the forwarding plane and into a centralised controller, but a physically centralised controller can represent a single point of failure for the entire network. This centralisation approach brings optimality, however, it creates additional problems of its own including single-domain restriction, scalability, robustness and the ability for switches to adapt well to changes in local environments. This research aims at developing a new distributed active information model (DAIM) to allow programmability of network elements and local decision-making processes that will essentially contribute to complex distributed networks. DAIM offers adaptation algorithms embedded with intelligent information objects to be applied to such complex systems. By applying the DAIM model and these adaptation algorithms, managing complex systems in any distributed network environment can become scalable, adaptable and robust. The DAIM model is integrated into the SDN architecture at the level of switches to provide a logically distributed control plane that can manage the flow setups. The proposal moves the computational load to the switches, which allows them to adapt dynamically according to real-time demands and needs. The DAIM model can enhance information objects and network devices to make their local decisions through its active performance, and thus significantly reduce the workload of a centralised SDN/OpenFlow controller. In addition to the introduction (Chapter 1) and the comprehensive literature reviews (Chapter 2), the first part of this dissertation (Chapter 3) presents the theoretical foundation for the rest of the dissertation. This foundation is comprised of the logically distributed control plane for SDN networks, an efficient DAIM model framework inspired by the O:MIB and hybrid O:XML semantics, as well as the necessary architecture to aggregate the distribution of network information. The details of the DAIM model including design, structure and packet forwarding process are also described. The DAIM software specification and its implementation are demonstrated in the second part of the thesis (Chapter 4). The DAIM model is developed in the C++ programming language using free and open source NetBeans IDE. In more detail, the three core modules that construct the DAIM ecosystem are discussed with some sample code reviews and flowchart diagrams of the implemented algorithms. To show DAIM’s feasibility, a small-size OpenFlow lab based on Raspberry Pi’s has been set up physically to check the compliance of the system with its purpose and functions. Various tasks and scenarios are demonstrated to verify the functionalities of DAIM such as executing a ping command, streaming media and transferring files between hosts. These scenarios are created based on OpenVswitch in a virtualised network using Mininet. The third part (Chapter 5) presents the performance evaluation of the DAIM model, which is defined by four characteristics: round-trip-time, throughput, latency and bandwidth. The ping command is used to measure the mean RTT between two IP hosts. The flow setup throughput and latency of the DAIM controller are measured by using Cbench. Also, Iperf is the tool used to measure the available bandwidth of the network. The performance of the distributed DAIM model has been tested and good results are reported when compared with current OpenFlow controllers including NOX, POX and NOX-MT. The comparisons reveal that DAIM can outperform both NOX and POX controllers. The DAIM’s performance in a physical OpenFlow test lab and other parameters that can affect the performance evaluation are also discussed. Because decentralisation is an essential element of autonomic systems, building a distributed computing environment by DAIM can consequently enable the development of autonomic management strategies. The experiment results show the DAIM model can be one of the architectural approaches to creating the autonomic service management for SDN. The DAIM model can be utilised to investigate the functionalities required by the autonomic networking within the ACNs community. This efficient DAIM model can be further applied to enable adaptability and autonomy to other distributed networks such as WSNs, P2P and Ad-Hoc sensor networks.
Please use this identifier to cite or link to this item: