DEVELOPMENT THE RESOURCES LOAD VARIATION FORECASTING METHOD WITHIN CLOUD COMPUTING SYSTEMS

• Published in: Сучасні інформаційні системи Vol. 4; no. 4; pp. 128 - 135
• Main Authors: Davydov, Viacheslav, Hrebeniuk, Daryna
• Format: Journal Article
• Language: English
• Published: National Technical University "Kharkiv Polytechnic Institute" 14.12.2020
• Subjects: cloud computing systems; Elman neural network; resources load forecasting
• ISSN: 2522-9052
• DOI: 10.20998/2522-9052.2020.4.18

The subject of research in the article is the models and methods of resources load forecasting in cloud computing systems using the mathematical apparatus of neural networks. The aim of the work is increasing the efficiency of computing systems resources usage (such as RAM, disk space, CPU, network) by developing methods of resources load forecasting. The article addresses the followingtasks: development of an integrated approach to the problems of resources load forecasting within cloud computing systems, which includes the synthesis of a combining forecasting neural network; development of a forecasting neural network model based on Elman neural network; development of a method for training a neural network based on an artificial immunity algorithm; evaluation of the effectiveness of the developed method. To solve the set tasks, the approaches and methods of artificial neural and immune systems were used, as well as methods of theoretical research, which are based on the scientific provisions of the theory of artificial intelligence, statistic, functional and systemic analyzes. The following results were obtained: on the basis of the analysis of resources load forecasting methods in cloud computing systems, the main results of the methods were revealed, the advantages and disadvantages were demonstrated. On the basis of the research results analysis, the necessity of improving analytical methods for forecasting the load has been proved. The method of computing resources load forecasting in cloud computing systems has been improved, which makes it possible to obtain more accurate assessment results and prevent overloads in cloud computing systems. The results obtained are confirmed by the experiments carried out using the means of the infrastructure of private infrastructure services. Conclusions: improved the resources load forecasting method based on the mathematical apparatus of artificial neural networks to improve the efficiency of their usage.