ue 16(11): e2

Research Article

Routing policies for a partially observable two-server queueing system

Download1143 downloads
  • @ARTICLE{10.4108/eai.14-12-2015.2262697,
        author={Wendy Ellens and Peter Kovacs and Rudesindo N\^{u}\`{o}ez-Queija and Hans van den Berg},
        title={Routing policies for a partially observable two-server queueing system},
        journal={EAI Endorsed Transactions on Future Internet},
        keywords={routing, tandem queue, dynamic control, incomplete information, partial control, fluid approximation},
  • Wendy Ellens
    Peter Kovacs
    Rudesindo Núñez-Queija
    Hans van den Berg
    Year: 2016
    Routing policies for a partially observable two-server queueing system
    DOI: 10.4108/eai.14-12-2015.2262697
Wendy Ellens1, Peter Kovacs1,*, Rudesindo Núñez-Queija1, Hans van den Berg2
  • 1: University of Amsterdam, Korteweg - de Vries Institute for Mathematics
  • 2: TNO / University of Twente
*Contact email: p.kovacs@uva.nl


We consider a queueing system controlled by decisions based on partial state information. The motivation for this work stems from road traffic, in which drivers may, or may not, be subscribed to a smartphone application for dynamic route planning. Our model consists of two queues with independent ex-ponential service times, serving two types of jobs. Arrivals occur according to a Poisson process; a fraction of the jobs (type X) is observable and controllable. At all times the number of X jobs in each queue and their individual po-sitions are known. Upon its arrival a router decides which queue the next X job should join. Y jobs are non-observable and non-controllable. They randomly join a queue according to some static routing probability. We address the following main research questions: 1) what penetration level is needed for effective control, 2) which policy should be implemented at the router, and 3) what is the added value of having more system information (e.g., average service times)? An extensive simulation study re-veals that for heavily loaded systems a low penetration level suucces and that the performance (in terms of the average sojourn time) of a simple policy that relies on little system information is close to w-JSQ (weighted join-the-shortest- queue policy) which is optimal in a fully controllable and observable system. The latter result is confirmed by the analysis of deterministic uid models that approximate the stochastic evolution under large loads.