Approximating class-departure variability in tandem queues with downtime events: Regression-based variability function. (December 2017)

• Record Type: Journal Article
• Authors: Sagron, Ruth; Rabinowitz, Gad; Tirkel, Israel
• Is Part Of: Computers & operations research. Volume 88(2017)
• Page Start: 161
• Page End: 174
• Publication Date: 2017-12
• Subjects: Regression-based variability function (RBVF) -- Tandem queues -- Class-departure variability -- Queue performance -- Decomposition approximation methods; Operations research -- Periodicals; Electronic digital computers -- Periodicals; 004.05
• DOI: 10.1016/j.cor.2017.07.003
• Languages: English
• ISSNs: 0305-0548

Abstract

Highlights: We deal with performance measure's prediction in tandem queues with downtimes. The prediction is obtained by a proper approximation of class-departure variability. Our approximation based on decomposition and variability function method. We significantly reduce simulation efforts and improve accuracy, as well. Abstract: Predicting queue performance by approximating class-departure variability in tandem queues with downtime events via existing decomposition methods is neither accurate enough nor efficient enough. Analytic approximations, if conducted alone, lack accuracy but attempting to increase accuracy by incorporating simulation to analytic approximation has proved to require significant computation efforts. The aim of this paper is to reduce the latter inefficiency by modeling the Regression-Based Variability Function (RBVF) designed to approximate the between-class effect by exploiting the departure process from a single queue. The new approach predicts performance of n -tandem queues by reducing the focus to two-tandem queues for each traffic intensity level, as well as by modeling different policies of downtimes (e.g. first-come-first-served or priority). Numerical experiments demonstrate that the proposed RBVF delivers both accuracy and efficiency improvements: the relative errors associated with RBVF are about three times smaller than the best existing analytic procedures and the computation efforts associated with RBVF are about five times smaller than existing analytic procedure combined with simulation.