Search academic texts

Information × Registration Number 0211U006413, 0108U001959 , R & D reports Title Development of theory of synthesis and distributed simulation of high performance problem-oriented computing systems with programmable VLSI circuits popup.stage_title Head Bojun Vitaly Petrovich, Registration Date 23-03-2011 Organization The state higher educational institution " Donetsk National Technical University" popup.description2 Report: 195 p., 130 images, 18 tables, 89 sources. The object of research is synthesis and distributed simulation of the high-performance problem-oriented processors and computing systems with programmable VLSI circuits. The goal of the work is development of new methods for design and simulation of problem-oriented processors and computing systems with programmable VLSI circuits for solving systems of linear equations, processing and control of data flows in real time. Research methods are systematization, generalization, mathematical modeling and computing experiment. Specialized processors on PLD for solving linear systems oriented to classic direct and iterative methods are analyzed. Speed of convergence and numerical stability of iteration incremental algorithms for solving linear systems are experimentally evaluated. The architecture of the specialized FPGA-based multiprocessor for the solving linear systems by incremental methods is proposed. A mapping of the video object tracking method DSOFT to parallel computing systems with MIMD architecture is developed; the performance and optimal load balancing are estimated analytically and experimentally. Hardware-oriented algorithms for search of arbitrary shape image fragments in a frame for multithread graphic processors with CUDA architecture are developed based on per-pixel comparison of frame and template, acceleration of search compared to CPU is experimentally evaluated. The performance of the three structures of network processor (with a common block of RAM, with separate memory, with dedicated instruction cache) is evaluated; their comparative analysis is made. A simulation system of networks-on-chip is developed. A model of hardware implementation of AntNet routing algorithm for a network-on-chip is developed and experimentally evaluated compared to the XY routing algorithm. A method of local cascade rollbacks for simulation of routing algorithms that takes into account heterogeneity of threads of events in logical processes is developed and experimentally evaluated on the closed network with queues. Algorithms of the dynamic balancing of loading for the distributed logical simulation are proposed those minimize migration of tasks, their experimental research is fulfilled. Results of research are used in the development of systems of distributed processing and simulation and in the educational process of university. SYSTEMS OF LINEAR EQUATIONS, SPECIALIZED PROCESSORS, INCREMENTAL ALGORITHMS, FPGA, TRACKING OF OBJECTS IN VIDEOS, PARALLEL ALGORITHMS, MIMD, SEARCH OF IMAGE FRAGMENTS, CUDA, NETWORK PROCESSOR, ANTNET, NETWORK ON CHIP, ROUTING, CASCADE ROLLBACKS, LOAD BALANING, MIGRATION Product Description popup.authors Бельков Д.В. Гомозов О.В. Грищенко В.І. Куркчі В.А. Ладиженський Ю.В. Мерецька В.А. Мерецький О.В. Середа А.О. popup.nrat_date 2020-04-02 Close