The thesis consists of an introduction, 4 chapters, conclusions, a list of references, and 5 appendices. The total volume of work is 172 pages, of which 118 pages are the main text, 35 figures, 9 tables.
Ph.D. thesis is devoted to the development of methods and tools for researching the influence of parallel computing parameters on the speed of computing. Modern information technologies require fast operation of algorithms, which can be achieved through the use of parallel computing. However, depending on the parameters that determine the characteristics of the subtasks and the mechanisms of their interaction, the use of parallel computing can lead to both speeding up and slowing down the computing. Mathematical methods for evaluating the efficiency of parallel computing are able to indicate the existence of a limit on the maximum achievable speed up under ideal conditions of free access to the computing resource and the absence of synchronization of calculations. Existing program design tools, such as UML, allow (quite generalized) to graphically represent the interaction of individual parts of the program but do not provide any numerical analysis of calculations. To date, there is experience in the development of Petri net-based computing simulators, but none of them have become widely used in the development of parallel computing. Thus, today, there is no unified method for creating a model of parallel computing and, accordingly, there is no tool other than a real program that can be used to optimize the parameters of a parallel program. In view of this, the creation of methods and tools aimed at improving the process of setting up multi-threaded programs and increasing the efficiency of using parallel computing in information technology is an urgent scientific task.
The purpose of scientific research is to increase the effectiveness of parallel computing usage in information technologies by designing it based on models which can be used for performance time estimation of parallel algorithm and parallel computing parameters optimization.
The object of research is the process of developing parallel computing in information technologies.
The subject of research is methods and means of modeling and optimization of parallel computing in information technologies.
Scientific novelty of the obtained results:
- for the first time, а parallel computing simulation technology based on the Petri-object approach has been developed, which provides an opportunity to reduce resource costs in the development of parallel algorithms. Unlike existing ones, it allows to reproduce in detail the structure of a parallel program and the mechanisms of interaction of simultaneously executing parts of the program, taking into account time delays on the execution of computing actions and the stochasticity of capturing the computing resource. Developed technology simplifies the process of building a model due to the replication of program fragments with similar functionality;
- the models of the basic mechanisms of synchronization of parallel computing have been improved by increasing the accuracy of reproduction, which ensures high accuracy of simulation results;
- for the first time, typical fragments of Petri nets implementing the mechanisms of Java multithreading technology have been developed, the use of which accelerates the development of a parallel algorithm model by reducing the number of errors and reducing the total number of elements required for model development;
- for the first time, a technology for optimizing the parameters of parallel computing based on an experimental study of the Petri-object model of computing is proposed. It ensures the efficient use of computing resources and, unlike existing approaches, makes it possible to conduct an experimental study of the efficiency of parallel computing on a model instead of experimenting on a real program.
The practical significance of the results of the dissertation research lies in the developed software for modeling and optimizing the parameters of parallel computing based on Petri-object simulation.
The results of the dissertation were published in 9 scientific publications, including 3 papers in a periodical scientific publication indexed in the Web of Science Core Collection and Scopus databases, 2 papers in a professional scientific journal, 3 publications in the materials of international scientific conferences indexed in Scopus, 1 publication in the materials of the All-Ukrainian scientific conference.
Keywords: multithreaded programming, parallel computing, optimization, simulation, stochastic Petri net, Petri-object simulation.
