Spring 2011 Seminar Series

DATE: April 21st
LOCATION: E62-550
TIME: 4:15pm
Reception immediately following in the same room

SPEAKER:
Retsef Levi

TITLE
Near-Optimal LP-Based Policies for Revenue Management of Reusable Resources

ABSTRACT
We consider a class of core models for revenue management of reusable resources that capture several important application domains. The main challenge in these models is to dynamically manage the allocation of capacitated reusable resources in the face of stochastic arrivals of multiclass customers, whereas the goal is to maximize the expected revenue. The resulting models are computationally challenging, even for very special cases. Based on linear programming relaxations of the models, we devise conceptually and computationally simple polices that are called class selection policies (CSPs). Moreover, we show that these policies have provably uniform and asymptotic worst-case guarantees in several asymptotic parametric regimes. The analysis is based on the well-studied stochastic models of loss-networks that arise in many applications. As part of the analysis, we obtain several new fundamental results on these important models, specifically on the blocking probabilities (i.e., the long-run probability that a random customer will be denied because of lack of capacity).

 

The talk is based on two papers jointly (respectively) with Ana Radovanovic and Cong Shi.