The paper presents a taxonomy for parallel processing systems. Because of technological limitations, high-speed computing has to be achieved nowadays by innovative architectures whose distinctive features cannot be cl...
详细信息
The paper presents a taxonomy for parallel processing systems. Because of technological limitations, high-speed computing has to be achieved nowadays by innovative architectures whose distinctive features cannot be clearly brought out by any of the taxonomies proposed to date. The paper discusses the drawbacks of these taxonomies before proposing a new method of classification based on some dominant architectural features, such as the level of concurrency, method of algorithm realisation, execution characteristic and control structure. The principal aim of proposing a new taxonomy is to develop a nomenclature which can be used to describe the salient features of all the types of parallel processing systems that have been proposed to date. In the last part of the paper it is shown how the nomenclature suggested in the proposed scheme of classification can be used to describe the ‘pure’ high-speed computing systems such as systolic array, pipeline processor, multiprocessor, data-flow processor, multifunction processor, array processor, associative processor and also the systems formed by hybridation of these ‘pure’ systems.
In this paper, a parallel computing system based on a data-flow model is applied to signal processing. A prototype data-flow processor is presented and the recursive digital filter is implemented on it as a typical ex...
详细信息
In this paper, a parallel computing system based on a data-flow model is applied to signal processing. A prototype data-flow processor is presented and the recursive digital filter is implemented on it as a typical example of signal processing. Since the effective I/O interface and the real-time processing of data sequence are essentially important in signal processing, stream-oriented processing is adopted to treat I/O and to set up the pipeline processing. An architectural extension which allows the exploitation of parallelism not only at the process level but also at the instruction level within the process is discussed.
暂无评论