Stirenko S. Methods and tools for efficient processing of parallel tasks in computer cluster systems

Dissertation is devoted to the theory of parallel computing cluster systems by developing new efficient methods and tools accelerate computational processes with regular and irregular parallelism, more efficient use of computing resources and memory. The concept of extension of the standard model of processing tasks in computing cluster systems containing hardware and software interaction, control and automatic balancing parallel tasks in a dynamic mode that allows to use the potentiality of improving the efficiency of the computational process. New methods of data exchange in parallel systems with local memory, in comparison with known methods can improve the performance of systems by reducing the time required to synchronize data exchange processes. Methods and tools for automatic dynamic parallelizing computations level subtasks, software modules and applications was proposed for cluster systems with local memory to reduce the total execution time of computation and increase the degree of loading equipment.