Methods for Accelerating the Search for the Optimal Distribution of Parallel Processes in Heterogeneous Systems

The paper considers the problem of searching the optimal distribution of tasks available for parallel execution among the operating units of a heterogeneous computing system. This problem generates an exponentially growing set of possible solutions the analysis of which takes essential time. In this paper we propose methods to reduce search space based on task decomposition, analysis of the code and architecture of heterogeneous system to accelerate the process of solving of optimization problem.