Thursday, May 01, 2014, 02:30pm
Graduate Student, EE Dept, University of Southern California
"Transitory Queueing Theory"
Abstract: Queueing Theory has historically focused on the analysis of stationary and ergodic systems. In the last couple of decades there has been much theoretical progress in analyzing time-varying systems. In this talk, we make the case that there are non-stationary and (possibly) non-ergodic systems that even existing theory does not address adequately. We identify such systems, demonstrate that discrete event analysis is difficult, if not impossible, and develop well justified approximations to the performance metrics of these models.
The systems of interest to us are those that either have a finite population of customers to serve, or only operate for a finite interval of time. We call these Transitory Queueing Systems. We introduce the notion of aTransitory Queueing Model, and the Population Acceleration (PA) technique, and prove diffusion and fluid approximations to a generic transitory model. We demonstrate that the class of models that fit this paradigm is broad, by studying the so-called ?(i)/GI/1 queue (a model of queueing we introduce), a multi-server queue with traffic modeled as a renewal process conditioned on a specific event of interest, and a model of queueing with scheduled arrivals that display uncertainty in the realized arrival epochs. We show that, even though these systems are radically different in the pre-limit description, the limiting behavior is (surprisingly) similar. Applications of this work include systems in retail, transportation, hospitals/clinics and even call centers.
Bio: Harsha Honnappa is a Ph.D. candidate and Ming Hsieh Institute Scholar in the Department of Electrical Engineering at USC, where he works with Prof. Rahul Jain of EE and Prof. Amy R. Ward of the Marshall School of Business. His research interests include stochastic networks, applied probability, statistics and game theory/network economics. His dissertation work has focused on the game theoretic modeling of traffic in single-class queueing networks, as well as the development of new analytical techniques for studying non-stationary and non-ergodic queueing systems.
Hosted by: EECS Prof. Vijay Subramanian