ABSTRACT
Computer architectures have passed through many notable evolutionary changes over the last few decades, yielding many different forms of architectural designs that are categorized according to Flynn’s classification of computer architectures: SISD, SIMD, MISD, and MIMD. The introduction of parallel computers (mainly MIMD) together with the required environment, and the different types of interconnection networks (static and dynamic) used to connect different resources in this type of systems have been briefly described. One form of MIMD, tightly-coupled multiprocessors (Shared-memory UMA model, Symmetric multiprocessor (SMP)) as parallel processing systems, and loosely-coupled multiprocessors (distributed-shared memory, NUMA model) and its commercially popular variants (CC-NUMA model) as distributed computing systems have been described. The other form of MIMD is multicomputers, essentially a network of computers having different types of architectural models including computer networks and distributed systems. This form of MIMD has been further enhanced to construct massively parallel distributed-memory machines, large multicomputer configurations, clusters of computers as well as massively parallel processing systems and supercomputers using various types of static as well as dynamic interconnection networks having different topologies. The cluster architecture of computers with its classification and different methods of clustering has been explained, including the role of the operating system that drives it. The introduction of blade servers and their enormous impact in large-scale distributed systems built with cluster architecture have been illustrated. With the proliferation of a diverse spectrum of application areas, certain classes of typically special-purpose problems have evolved, which require special types of processors known as SIMD processors (according to Flynn’s definition), which again is classified into two broad categories – Vector processors, and Array processors – offering more fine-grained data parallelism. SIMD computer organizations including SIMD vector processors and SIMD array processors have been discussed with related illustrations and usage. Massively parallel processing (MPP) systems that provide both hardware parallelism and software parallelism have been ultimately realized by various other means to meet the forthcoming challenges of computing environments. The supercomputer of today is essentially a massively parallel processing system, and most modern supercomputers are now highly--tuned computer clusters using commodity processors combined with fast custom-interconnects.
