In spite of the wealth Of existing data distribution methods, most parallel programming languages support only some form of cyclic blockwise distribution. The main reason why only this single method is supported is th...In spite of the wealth Of existing data distribution methods, most parallel programming languages support only some form of cyclic blockwise distribution. The main reason why only this single method is supported is that it is relatively simple to implement. However, it is as yet nuclear whether cyclic blockwise distribution is sufficiently powerful for a wide class of distribution problems. In this paper the method will be analysed, showing that for a wide range of problems it is indeed sufficient. It will also be shown in which cases cyclic blockwise distribution can be effected to fail. From this analysis, it is possible to formulate practical guidelines to assist Programmers in choosing the cycle frequency for cyclic blockwise distribution that leads to an optimal result.展开更多
文摘In spite of the wealth Of existing data distribution methods, most parallel programming languages support only some form of cyclic blockwise distribution. The main reason why only this single method is supported is that it is relatively simple to implement. However, it is as yet nuclear whether cyclic blockwise distribution is sufficiently powerful for a wide class of distribution problems. In this paper the method will be analysed, showing that for a wide range of problems it is indeed sufficient. It will also be shown in which cases cyclic blockwise distribution can be effected to fail. From this analysis, it is possible to formulate practical guidelines to assist Programmers in choosing the cycle frequency for cyclic blockwise distribution that leads to an optimal result.