Simulation modeling and optimization of the operation of a parallel server with failures in AnyLogic

About the authors:

Lyudmila B. Senkevich, Cand. Sci. (Ped.), Associate Professor, Department of Cybernetic Systems, Tyumen Industrial University;  lyudmila1@yandex.ru

Marat A. Sabitov, Master Student, Department of Cybernetic Systems, Tyumen Industrial University;  sabitov.m.a@yandex.ru

Abstract:

Modern scientific research is increasingly raising issues of the processing of large amounts of data. The widespread use of client-server interaction technology and cloud computing at the moment raises questions about the efficiency of a parallel server, as well as the ability to predict results depending on the degree of load and characteristics of the equipment.

This article simulates a parallel server with failures in the AnyLogic environment, and then performs multidimensional optimization by the weighted sum method. As part of the study, a simulation model of a queuing system with failures was built. It contains a server simulator, terminals, a failure simulator and statistics collection segments. The used parallel server model is abstract and rather generalized and makes it possible to concretize it by introducing additional dependencies and refining characteristics. The experiment with optimal parameters allowed to obtain the following gain in system efficiency indicators: processor load parameter (by memory) — a gain of 7%; processor load parameter (by load factor) — a gain of 8%; probability of terminal downtime — a gain of 5.7%; the failure rate of the main computer — 36 times less than the initial configuration; the number of interrupted programs — 7 less. In addition, it should be noted that the total number of completed requests remained at the same level — 462-465, for the reason that the intensity of the terminals did not vary.

Since the results of replications (“runs”) are unique and the values of the optimized function vary for different replications, the built-in possibility of a variable number of replications (from 5 to 10) with a confidence probability of 95% and an error level of 0.5 was used. The obtained results suggest the possibility of further research of the model and its development in the AnyLogic environment.

References:

1. Kislitsyn E. V., Pershin V. K. 2013. System modeling: discrete-event approach. Ekaterinburg: USUE. 101 p. [In Russian]

2. Makarov V. L., Bakhtizin A. R., Beklaryan G. L. 2019. “Development of digital twins for manufacturing enterprises”. BusinessInformatics, no. 13 (4), pp. 7-16. [In Russian]

3. Obukhov P. A., Nikolaev A. B., Ostroukh A. V. 2015. “Study of the efficiency of network servers in the AnyLogic simulation environment”. International Journal of Experimental Education, no. 3-3, pp. 338-342. [In Russian]

4. Optimization experiment. Accessed on 11 October 2021. https://help.anylogic.ru/index.jsp?topic=%2Fcom.anylogic.help%2Fhtml%2Fexperiments%2Foptimization-ex... [In Russian]

5. Rzhevskiy S. V. 2013. Operation Research. Saint-Petersburg: Lan. 480 p. [In Russian]

6. Rott A. R. 2010. Modeling and calculations of production and technical systems. Yoshkar-Ola: Volga State University of Technology. 224 p. [In Russian]

7. Steen van М., Tanenbaum A. S. 2021. Distributed systems. Мoscow: DMK Press. 584 p. [In Russian]

8. The Monte Carlo Experiment. Accessed on 11 October 2021. https://help.anylogic.ru/index.jsp?topic=%2Fcom.anylogic.help%2Fhtml%2Fexperiments%2Fmonte-carlo-exp... [In Russian]

9. Arora R. 2015. Optimization: algorithms and applications. Boca Raton: Chapman and Hall/CRC. 466 p.

10. Luc D. T. 2016. Multiobjective linear programming: an introduction. London: Springer. XII, 325 p. DOI: 10.1007/978-3-319-21091-9