Design and analysis of a 3-dimensional cluster multicomputer architecture using optical interconnection for petaFLOP computing

In this dissertation, the design and analyses of an extremely scalable distributed multicomputer architecture, using optical interconnects, that has the potential to deliver in the order of petaFLOP performance is presented in detail. The design takes advantage of optical technologies, harnessing the features inherent in optics, to produce a 3D stack that implements efficiently a large, fully connected system of nodes forming a true 3D architecture. To adopt optics in large-scale multiprocessor cluster systems, efficient routing and scheduling techniques are needed. To this end, novel self-routing strategies for all-optical packet switched networks and on-line scheduling methods that can result in collision free communication and achieve real time operation in high-speed multiprocessor systems are proposed. The system is designed to allow failed/faulty nodes to stay in place without appreciable performance degradation. The approach is to develop a dynamic communication environment that will be able to effectively adapt and evolve with a high density of missing units or nodes. A joint CPU/bandwidth controller that maximizes the resource allocation in this dynamic computing environment is introduced with an objective to optimize the distributed cluster architecture, preventing performance/system degradation in the presence of failed/faulty nodes. A thorough analysis, feasibility study and description of the characteristics of a 3-Dimensional multicomputer system capable of achieving 100 teraFLOP performance is discussed in detail. Included in this dissertation is throughput analysis of the routing schemes, using methods from discrete-time queuing systems and computer simulation results for the different proposed algorithms. A prototype of the 3D architecture proposed is built and a test bed developed to obtain experimental results to further prove the feasibility of the design, validate initial assumptions, algorithms, simulations and the optimized distributed resource allocation scheme. Finally, as a prelude to further research, an efficient data routing strategy for highly scalable distributed mobile multiprocessor networks is introduced.