INFORMATION TECHNOLOGY FOR DETERMINING USEFUL DATA WHILE OPTIMIZING THE STRUCTURE AND MINIMIZING THE VOLUME OF THE DISTRIBUTED DATABASE NODE

• Main Authors: Михайло Леонідович Дворецький, Світлана Володимирівна Дворецька, Євген Олександрович Давиденко
• Format: Article
• Language: English
• Published: Cherkasy State Technological University 2020-01-01
• Series: Вісник Черкаського державного технологічного університету
• Subjects: розподілена транзакція; система керування базами даних; розподілена база даних; розподілений sql-запит; реплікація даних; парсинг тексту; дерево парсингу; профайлінг; antlr; olap; задача класифікації; нейронна мережа; інтелектуальний аналіз даних.
• Online Access: http://vtn.chdtu.edu.ua/article/view/184808
• ISSN: 2306-4412; 2708-6070

The paper deals with the tendency to move from "universal" accounting systems to specialized solutions usage. This requires the synchronization of distributed database data. It is noted that among the strategies of data distribution between distributed database nodes, the combined one is the most justified, but the main disadvantage consists in the existence of distributed transactions when handling data. The research aims to improve the general availability of data in the separate node of the distributed database and the efficiency of using software systems to work with database data by reducing the number of distributed requests. The goal is achieved by optimizing the structure of the distributed database node and minimizing the amount of data stored in it. To achieve the goal, users' query accounting subsystem and T-SQL grammar have been created, and SQL query code has been parsed. As a result, the queries are classified by the list of database tables that are found in the query, and, after performing more deyailed analysis, by the list of attributes and relation tuples. The last one is achieved by executing a set of queries with getting the primary key of each relation included in the query. Performing the complete analysis of the database tables attributes and tuples estimation is a very resource-intensive operation, so it cannot be performed every time the database data is changed. The research proposes to solve the problem of classification of new data by using the perceptron, which learns on the basis of pre-evaluated data based on SQL query parsing. Also, according to the need of performing the analysis of received data from the point of view of multiple dimensions, as well as probably their large amount, the data required for the analysis has been presented in the form of a multidimensional model.