CASPIAN JOURNAL
MANAGEMENT AND HIGH TECHNOLOGIES
Reducing the approximation time of cluster workload by using method of moments on hyperexponential distribution
 |
Read | Gaevoy S.V., Akhmed Vesam Mokhammed Ali, Bykov D.V., Fomenkov S.A. Reducing the approximation time of cluster workload by using method of moments on hyperexponential distribution // Caspian journal : management and high technologies. — 2017. — №1. — pp. 94-105. |
Gaevoy S.V. - Ph.D. (Engineering), Volgograd State Technical University, 28 Lenina Ave., Volgograd, 400005, Russian Federation, gaevserge@mail.ru
Akhmed Vesam Mokhammed Ali - postgraduate student, Volgograd State Technical University, 28 Lenina Ave., Volgograd, 400005, Russian Federation, wesamalsofi@gmail.com
Bykov D.V. - Ph.D. (Engineering), Associate Professor, Volgograd State Technical University, 28 Lenina Ave., Volgograd, 400005, Russian Federation, mitril@list.ru
Fomenkov S.A. - D.Sc. (Engineering), Professor, Volgograd State Technical University, 28 Lenina Ave., Volgograd, 400005, Russian Federation, saf@vstu.ru
Computing clusters are one of the computing systems. They are used to execute incoming jobs. An important method to analyze parallel workloads is modeling execution of those systems by using parallel workload models. In this paper it is proposed to use method of moments to compute parameters of Hyperexponential distribution with two branches and get a parallel workload model. This allows us to drastically reduce the approximation time of the parallel workload model in comparison to maximum likelihood method, but it reduces the quality too. To validate the result quality we use the simulation of this approximation and compare the results with the original workload (from the log) in this paper. The results of the formerly proposed parallel workload models are compared with the results from this paper. The reasonability to select an appropriate appro ximation method for solving approximation tasks is justified.
Key words: метод моментов, интегральная функция распределения, нагрузки вычислительных систем, не масштабируемые задачи, время выполнения заданий, имитационное моделирование, стохастическая аппроксимация, method of moments, cumulative distribution function, parallel