Efficiency analysis of the parallelizing of population optimization methods by proxy-applications

The paper is devoted to the model and its software implementation for automatic parallelizing of evolutionary and swarm optimization algorithms. The paper considers the classification of interaction patterns between individuals of a population, on the basis of which a method of automatic parallelization of algorithms of this class is proposed, taking into account various models of parallel execution. It is shown how, within the framework of the proposed model, it is possible to build parameterized proxy-applications that emulate the parallel execution of a given algorithm, in order to study the parallelization efficiency, as well as for optimal tuning of the hardware.