Efficient simulation of a tandem queue with server slow-down


Miretskiy, D.I. and Scheinhardt, W.R.W. and Mandjes, M.R.H. (2006) Efficient simulation of a tandem queue with server slow-down. In: 6th International Workshop on Rare Event Simulation, 8-10 October 2006, Bamberg, Germany (pp. pp. 132-143).

open access
Abstract:Tandem Jackson networks, and more sophisticated variants, have found widespread application in various domains. One such a variant is the tandem queue with server slow-down, in which the server of the upstream queue reduces its service speed as soon as the downstream queue exceeds some prespecified threshold, to provide the downstream queue some sort of `protection'.

This paper focuses on the overflow probabilities in the downstream queue. Owing to the Markov structure, these can be solved numerically, but the resulting system of linear equations is usually large. An attractive alternative could be to resort to simulation, but this approach is cumbersome due to the rarity of the event under consideration. A powerful remedy is to use importance sampling, i.e., simulation under an alternative measure, where unbiasedness of the estimator is retrieved by weighing the observations by appropriate likelihood ratios.

To find a good alternative measure, we first identify the most likely path to overflow. For the normal tandem queue (i.e., no slow-down), this path was known, but we develop an appealing novel heuristic, which can also be applied to the model with slow-down. Then the knowledge of the most likely path is used to devise importance sampling algorithms, both for the normal tandem system and for the system with slow-down. Our experiments indicate that the corresponding new measure is sometimes asymptotically optimal, and sometimes not. We systematically analyze the cases that may occur.
Item Type:Conference or Workshop Item
Electrical Engineering, Mathematics and Computer Science (EEMCS)
Research Group:
Link to this item:http://purl.utwente.nl/publications/66689
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page

Metis ID: 237678