A use of hyperexponential distribution in non-markovian queuing systems analyses | Vestnik Tomskogo gosudarstvennogo universiteta. Upravlenie, vychislitelnaja tehnika i informatika – Tomsk State University Journal of Control and Computer Science. 2016. № 3(36).

A use of hyperexponential distribution in non-markovian queuing systems analyses

In this paper we consider the application of second order hyperexponential distribution (H2) with complex-type parameters for analysis of non-markovian queuing systems. This distribution relates to the phase-type one and allows showing non-markovian queuing systems states and the transitions between them as discrete Markov process with continuous time. The complementary cumulative H2 distribution function is F (t) = y1e ^ + y 2 e ^. Using of H2 distribution for queuing system calculations is reasoned by following reasons: - the possibility of saving the three initial moments of the original distribution that provides high accuracy in queuing system calculation; - much more compact (compared to the Erlang approximation) view transition diagrams and as a consequence necessity to calculate the probabilities of a smaller number of microstates for systems with low coefficients of variation of the service time or the intervals between customers; - simple calculating of complementary cumulative distribution function. Since the parameters of H2-distributions {j}, J = 1, 2, are interpreted as the probabilities of random select of one of two phases, most specialists in queuing theory considered only the case when these parameters are defined on a real interval [0; 1], which corresponds to the approximation of the distribution with a coefficient of variation и > 1. In this article, it is showed the possibilities of the H2-approximation in the case when the original distribution coefficient of variation и < 1. In this case parameters of H2 distribution are complex. More detailed analysis of H2-approximation of the original gamma distribution with a coefficient of variation и shows that: - if и > 1, then the parameters are real; - if 1 > и , then the parameters are real, but the paradox: one of the parameters {j}, J = 1, 2, is negative, and the other will exceed one; - equality и = is unacceptable (because corresponds to the Erlang distribution with consistent phase-change and, accordingly, can not be replaced by parallel); - when и <1lV2 , we have the complex parameters of H2 distribution. However, when calculating the queuing system with H2-approximation in the field of complex values of the parameters of its potential pathology manifests itself only in the intermediate results - in the probabilities of "fictitious" microstate-transition diagram, which split the "physical" state of queuing systems. At the summation of probabilities of microstates tiers of complex parts are annihilated and the result of the calculation - the probability of the number of customers in the system - becomes real. The paper compares the results of single-channel systems MIGI1 calculation invested by embedded Markov chain, which allows you to obtain an exact solution, with the results obtained by the fictitious phase using H2-approximation of non-exponential service time. Various initial distribution services - deterministic, gamma with shape parameters of 1.5 or 0.5, and even in the interval [0, 2] are considered. It is shown that the accuracy of the above-mentioned result is high enough. The maximum Kolmogorov distance was 0.002, and the relative error of the probability of rare events (about 10) did not exceed 15%. At the same time the quality of approximation in the field of complex parameters H2 distribution is out of question because density of H2-distribution in this case is negative and in general does not satisfy the requirements of the probability density function. Quality of approximations here should be assessed indirectly - through comparison of the distribution of the number of customers in the queuing system, obtained through H2-approximation, with the results obtained by alternative methods.

Download file
Counter downloads: 275

Keywords

случайные процессы, аппроксимация, гиперэкспоненциальное распределение, комплексные параметры распределения, немарковские системы массового обслуживания, hyperexponential distribution, approximation, complex parameters of distributions, numerical methods, non-markovian queuing systems

Authors

NameOrganizationE-mail
Ryzhikov Yuriy I.Mozhaisky Military Space Academy; Saint-Petersburg Institute of Informatics and Automatization of Russian Academy of Scienceryzhbox@yandex.ru
Ulanov Alexander V.Mozhaisky Military Space Academyulanov246@rambler.ru
Всего: 2

References

Рыжиков Ю.И. Алгоритмический подход к задачам массового обслуживания. СПб. : ВКА, 2013. 496 с.
Рыжиков Ю.И., Уланов А.В. Применение гиперэкспоненциальной аппроксимации в задачах суммирования потоков // Ин теллектуальные технологии на транспорте. 2015. № 4. С. 34-39.
Рыжиков Ю.И., Уланов А.В. Расчет гиперэкспоненциальной системы M/H2/n-H2 с заявками, нетерпеливыми в очереди // Вестник Томского государственного университета. Управление, вычислительная техника и информатика. 2014. № 2(27). C. 47-53.
Цициашвили Г.Ш. Синергетический эффект в сети с гиперэкспоненциальными распределениями времен обслуживания // Вестник Томского государственного университета. Управление, вычислительная техника и информатика. 2016. № 1 (34). C. 65-68.
Назаров А.А., Бронер В.И. Система управления запасами с гиперэкспоненциальным распределением объемов потребления ресурсов // Вестник Томского государственного университета. Управление, вычислительная техника и информатика. 2016. № 1(34). C. 43-49.
Кендалл М.Дж., Стьюарт А. Теория распределений : пер. с англ. М. : Наука, 1966. 587 с.
Cox D.R. A Use of Complex Probabilities in the Theory of Stochastic Process // Proc. of the Cambridge Phil. Soc. 1955. P. 313-323.
Bidabad B., Bidabad B. Complex Probability and Markov Stochastic Process // Proc. of the First Iranian Statistics Conference, Isfa han University of Technology. 1992. P. 1-8.
Рыжиков Ю.И. Развитие и сопоставление методов расчета многоканальных систем обслуживания // Труды всероссийской конференции «XII Всероссийское совещание по проблемам управления» ВСПУ'2014 / Ин-т пробл. управл. М., 2014. С. 5208-5219.
 A use of hyperexponential distribution in non-markovian queuing systems analyses | Vestnik Tomskogo gosudarstvennogo universiteta. Upravlenie, vychislitelnaja tehnika i informatika – Tomsk State University Journal of Control and Computer Science. 2016. № 3(36).

A use of hyperexponential distribution in non-markovian queuing systems analyses | Vestnik Tomskogo gosudarstvennogo universiteta. Upravlenie, vychislitelnaja tehnika i informatika – Tomsk State University Journal of Control and Computer Science. 2016. № 3(36).

Download full-text version
Counter downloads: 769