Resumen
The article discusses the constructing of tasks dependencies graphs for many-task applications that perform parallel asynchronous data processing on the principle of round-robin sport tournament. The following components of the technique are described: the family of round robin algorithms is defined in the form of an algorithmic skeleton; a method for synthesizing a graph of a round-robin tournament using the basic sequential algorithm is proposed; it is shown how to use the graphs for the implementation of parallel and asynchronous computing process. A graphical interpretation of the round-robin tournament procedures is considered, on which it is possible to build analytical estimates of the tournament time. The technique is illustrated by three examples of constructing specific tournament algorithms: the tournament without additional restrictions, the simple sorting tournament, and the optimized sorting tournament. We build the algorithms for constructing task dependencies graphs for these three tournaments. The principle of implementing asynchronous parallel computing according to the graphs are considered. Using the graphical interpretation, the number of rounds and the speedup was calculated for each of the listed tournaments. The technique is recommended for a wide class of parallel architectures including clusters, desktop grids, and hybrid clouds.