Zhabina V. Methods and means for improvement of fine-grain parallelism realization effectiveness in real-time systems

Thesis deals with a problem of real-time fine-grain algorithm realization acceleration and its fault-tolerance assurance for computing systems designed on VLSI. Possibility to accelerate data processing in data-driven systems is proven. To achieve it, there is proposed realization of concurrent operations with computation units, which realize concurrent processing of operand places presented in redundant numerical system with large basic number. Method of data processing, which enables to accelerate computations at the cost of computation unit load intensity encrease by means of dynamic concurrent data flow generation, is developed for data-flow systems. Possibility to accelerate system recovery in case of computation unit fault is shown. Means of operation execution time control and system authomatic reconfigu-ration process are improved.