Optimization Algorithm for an Information Graph for an Amount of Communications

In connection with the annual increase in the volume of processed data and raising the importance of computer modeling of real objects and processes, requirement to improve the technology of parallel algorithms is increasing. Successful implementation of parallel algorithms on supercomputers depends on several parameters, one of which is the amount of inter-processor data transfers. Starting at a particular number of processors, computational speedup falls due to increased volume of data transmission. For some algorithms this dependence is a linear decreasing function. Imbalance of volume of calculations and complexity of data transmission operations increases with the rising of the number of processors. In this article we present the results of investigations of dependence of the density and algorithm execution time on the amount of interprocessor transfers. Also, we present a method of reducing interprocessor communications through more efficient distribution of operations of the algorithm by processes. This method does not account for the execution time of the operations themselves, but it is a foundation for more improved methods of multiparametric optimization of parallel algorithms.