A Paradox in a Queueing Network with State-Dependent Routing and Loss.pdfVIP

Hindawi Publishing Corporation Journal of Applied Mathematics and Decision Sciences Volume 2007, Article ID 68280, 10 pages doi:10.1155/2007/68280 Research Article A Paradox in a Queueing Network with State-Dependent Routing and Loss Ilze Ziedins Received 25 May 2007; Accepted 8 August 2007 Recommended by Paul Cowpertwait Consider a network of parallel finite tandem queues with two stages, where each arrival attempts to minimize its own cost due to loss. It is known that the user optimal and asymptotic system optimal policies may differ—we give examples showing that they may differ for finite systems and that as the service rate is increased at the second stage the user optimal policy may change in such a way that the total expected cost due to loss increases. Copyright © 2007 Ilze Ziedins. This is an open access article distributed under the Cre- ative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. 1. Introduction and background Queueing networks often exhibit seemingly paradoxical behaviour, where adding capac- ity either in the form of extra capacity on links or at nodes, or even extra links or routes, does not always lead to an improvement in performance, and may even lead to a severe degradation in performance. The classic and very well-known example of this is Braess’s paradox, which has been much studied, both in the traffic literature, and in the queueing theory literature (see [ 1, 2] for the original paper and [3] for a comprehensive list of ref- erences maintained by Braess). This was one of the examples mentioned by Professor Jeff Hunter in his inaugural lecture. It is therefore a great pleasure to write on a different kind of paradox in this festschrift for Jeff. In addition to Braess’s paradox, there are several other well-known paradoxes (see, e.g., Arnott and Small [4] and the references therein). The p