Study Of Cloud Service Oriented Optical Interconnection Network Technologies

Cloud computing is essentially to provide flexible information services to the users in a "pay-as-you-go" manner based on data centers (DC) and communication networks, which is just like water and electricity supply system During the last decade,Cloud services have developed so fast that traditional network infrastructures have faced more and more challenges to accommodate the dramatically increasing data communications inside/outside DCs. In the intra-DC scenario, the interconnections among tons of servers require higher scalability, higher throughput and lower latency, while conventional electrical packet switching intra-DC network architectures are not satisfying in all aspects. As for the metro networks which provide interconenctions among multiple DCs or between DCs and users, challenges also exist in how to make the networks more flexible and agile for service-differentiated bandwidth provisioning and intelligent failure recovery, etc. Under such circumstances, optical network technologies, expecially those multi-granular all-optical switching technologies, with distinguished high bandwidth and low energy consumption merits, have spawned interests of researchers worldwide and been regarded as among the most fundamental technologies for the future network infrastructure deployments. In this dissertation, we focus on the challenges and opportunities of Cloud service oriented optical networking technologies, dealing with the design and demonstrations of optical network architectures as well as the corresponding control and scheduling schemes for both intra-DC and inter-DC scenarios. Specifically, the contents of this dissertation are as follow:First, a novel intra-DC optical network architecture, which is named"OpenScale",is proposed and studied. The design principles of OpenScale architecture are introduced, which are basically inspired by the "small world topology"from complex network theories. We evaluated OpenScale network's performance by simulations. Simulation results show that, by exploiting the reconfiguration capability of optical switching technology, OpenScale network can achieve high throughput and low latency performance yet with prominent scalability at relatively low cost.Second, an LOBS-over-WSON network architecture, which features centralized control and scheduling, is proposed as a promising metro-network solution for accommodating data traffic brought by Cloud services. Specifically, we discuss the developing trend and requirements of future metro optical networks, and describe the LOBS-over-WSON solution which is an overlayed multi-granular optical switching network model and supports centralized end-to-end connection control and contention mitigation. Experimental demonstrations verified the feasibility of the proposed network architecture. Besides, as for a special case in which the network is built in the form of a synchronized ring, a novel protection switching scheme is proposed with the purpose of improving resource utilization after network failure. Simulation results proved that the proposed scheme performes better than the typical ring networks'failure recovery scheme.Third, the principles and methods of "software defined" unified control over heterogeneous resources for the optically interconnected Cloud infrastructures are studied. In regard to the intra-DC environment, we defined the coordinated scheduling relationships among applications, computing & storage resources and optical network resources. Based on an OpenScale system prototype, we experimentally demonstrated the proposed software defined control schemes which aim to dynamically map applications' requirements to intra-DC optical network resource scheduling. As for the inter-DC IP+Optical multilayer metro network scenario, we aslo studied the OpenFlow-based unified scheduling of multilayer/heterogeneous resources. Both simulations and experiments are conducted, and the results manifest the viability and effectiveness of our proposals.