Efficient, Scalable and Reliable Network (Function) Virtualization in Software-Defined Optical Networks
MetadataShow full item record
Software-Defined Networking (SDN) and Network Function Virtualization (NFV) are mutually beneficial and interactively complementary to provide an efficient way of deploying network services. SDN decouples the control plane from the data plane by abstracting the control intelligence into a centralized controller while leaving behind a programmable network. NFV decouples network functions from the proprietary hardware by implementing the network functions in software that can be run on standardized high-volume servers/switches/storage. Therefore, with SDN and NFV, deploying network services only requires launching specific virtual machines in datacenters and establishing connections between them, which yields a lower cost and a faster deployment, compared to the traditional way of purchasing and installing customized physical network equipment. In addition, multiple virtual machines can share the resources on the same physical hardware, and one can dynamically "stretch/contract" virtual machines to accommodate time-varying demands, thus highly improving the physical resource utilization. Recently, efforts are going on to integrate optical transport within the SDN architecture to leverage the benefits of optical transmission, such as low latency, long distance, and high bandwidth transmission. Such a network is referred to as a Software-Defined Optical Network (SDON). In this thesis, we focus on SDON controller design, resource allocation and network (function) virtualization in optically interconnected telecommunication networks. In this thesis, we first studied the traffic grooming problem for point-to-point traffic in both fixed and flexible grid optical networks. We proposed effective schemes that can make efficient grooming decisions to maximize the network resource utilization for both static and dynamic traffic. Furthermore, we studied the virtual infrastructure (VI) mapping problem that addresses not only the point-to-point traffic engineering for networking resources, but also the joint allocation for computing resources. We proposed effective algorithms to solve the cost-efficient VI mapping, upgrade-aware VI mapping and survivable VI mapping problems with the objective of minimizing the request blocking probability, while taking into consideration the physical layer constraints such as the transmission reach constraint and the wavelength/spectrum continuity constraint. Finally, we studied how to jointly optimize the topology design and mapping of multiple service function chains (SFCs), which is called the Joint Topology Design and Mapping (JTDM) problem. We proposed a set of algorithms to efficiently address the JTDM problem with the objective of minimizing the network resource consumption and the network reconfigurations, as well as maximizing the network service reliability. In general, the proposed strategies and algorithms in this thesis can be implemented and run as optimization applications in the SDON controller to achieve a cost-efficient, scalable and reliable network (function) virtualization in the next generation flexible grid optical networks.