Scientific journal
International Journal of Experimental Education
ISSN 2618–7159
ИФ РИНЦ = 0,425

1 1 1
1

The parallelization of computations for the high-productive supercomputer systems appears to be one of the main ways of development of the modern computational mathematics. The supercomputers are the more and more widely used for a solution of the fundamental and applied problems in the areas of nuclear physics, climatology, economics, pharmacology, modeling of the training devices, and of the virtual reality, computational aerodynamics. Due to those specific features of the Monte Carlo methods, which were repeatedly stressed in the present paper, the statistical modeling begins to play the more and more noticeable role in all, indicated above areas of science and techniques. For these reasons, the actuality of the problems mentioned is growing very considerably, taking into account the fact that the computational aerodynamics is the most promoted area of the elaboration, development, and application of the Monte Carlo methods [1]. As the mentioned above features of these methods permit to state, that the numerical schemes of a statistical modeling might be, in quite a natural way, transferred onto the parallel processors. Clearly, the successive modeling of the independent trajectories should be entrusted to the individual processors, while the information for the averaging will be gathered by a server [2]. In this case, the productivity of the method is growing in direct proportionality to the number of parallel processors.

Nowadays, as computer processors become cheaper and more plentiful, there is great potential for having them compute together in a coordinated application. A major point of parallel computing is how to coordinate communication between the various processors; indeed, some parallel computing techniques require specialized programming to permit the processors to work together in parallel. It can be seen that on Monte Carlo simulations, algorithms proceed by averaging large numbers of computed values. It is sometimes straightforward to have different processors compute different values, and then use an appropriate average of these values to produce a final answer.

The reported study was partially supported by the Russian Foundation for Basic Research (Research project No. 14-07-00564-а).


The work is submitted to the International Scientific Conference «Fundamental researches», Dominican Republic, April 13-22, 2014, came to the editorial office оn 21.03.2014