+7 (495) 987 43 74 ext. 3304
Join us -              
Рус   |   Eng

Authors

Efimenko A.

Degree
PhD in Technique, The Russian State University of Justice
E-mail
alex192@mail.ru
Location
Moscow
Articles

Optimization of the sizes of blocks of elementary tasks in problems of parallel computing planning

To solve problems parallel to the data, the computational complexes of the following architecture are gaining in popularity. The computing system consists of a universal part and m computational elements on SoC, FPGA, GPU, and so on. Computational elements receive tasks with blocks of size k, executing them in the future in an offline mode. The computing element may fail due to overheating or power failure. The probability of this event is known. After failure, we believe that the computational element is not restored. The larger the block size processed by one calculator, the less likely that the task will be performed on the remaining calculators. It is necessary to choose the size of the block in such a way as to obtain a guaranteed probability of completing the task. The article gives a equation that is suitable for calculations, which makes it possible to estimate the probability of accomplishing the task. This probability depends on the number of calculators and the size of blocks of elementary tasks. The results of a computational experiment for some values of computational elements and block sizes are presented. The equation allows you to select the optimal size of blocks of elementary tasks for a given probability of solving a problem.
Read more...