2015 Informs Annual Meeting

MA22

INFORMS Philadelphia – 2015

MA23 23-Franklin 13, Marriott Queues: Approximations and Control Sponsor: Applied Probability Sponsored Session Chair: Amy Ward, Professor, University of Southern California, Marshall School of Business, Los Angeles, CA, 90089, United States of America, amyward@marshall.usc.edu Co-Chair: Mor Armony, NYU Stern, 44 West 4th Street, New York, NY, 10012, United States of America, marmony@stern.nyu.edu 1 - Stein’s Method for Diffusion Approximations of Many-Server Queues Diffusion approximations for many-server queues have been studied extensively since the pioneering work of Halfin and Whitt (1981). We focus on many-server queues that have phase-type service time distributions and exponential customer patience distributions. We establish rate of convergence of the steady-steady distribution of a many-server queue in the Halfin-Whitt regime. Our proof technique connects naturally with Stein’s method for establishing normal approximations. 2 - Optimal Traffic Schedules Harsha Honnappa, Perdue University, West Lafayette, IN, United States of America, honnappa@purdue.edu, Rami Atar, Mor Armony We consider the problem of optimally scheduling a finite, but large, number of customers over a finite time horizon at a single server FIFO queue, in the presence of ‘no-shows’. We study the problem in a large population limiting regime as the number of customers scales to infinity and the appointment duration scales to zero. We show that in the fluid scaling heavy-traffic is obtained as a result of asymptotic optimization. We also characterize the diffusion-scale optimal schedule. 3 - Collaboration and Multitasking in Networks: Capacity Versus Queue Control Jan Van Mieghem, Professor, Kellogg School of Management, 2001 Sheridan Road, 5th Floor, Evanston, IL, 60201, United States of America, vanmieghem@kellogg.northwestern.edu, Itai Gurvich We study networks where some activities require the simultaneous processing by multiple types of multitasking human or indivisible resources. The capacity of such networks is generally smaller than the bottleneck capacity. This paper shows how this capacity is achieved through, and affected by, dynamic queue control. Prioritizing specific queues comes at a signicant loss of capacity. We present policies that balance capacity and queue control while guaranteeing stability and optimal scaling. 4 - Dynamic Scheduling in a Many-Server Multiclass System with General Abandonment Distributions Amy Ward, Professor, University of Southern California, Marshall School of Business, Los Angeles, CA, 90089, United States of America, amyward@marshall.usc.edu Our purpose is to understand how the assumed customer abandonment distribution affects scheduling decisions in a multiclass M/M/N+GI queue. To do this, we set up and solve an approximating diffusion control problem. We find that threshold control is in general sub-optimal. For two classes, we establish conditions for the optimal policy to have a “U-shape”, instead of the step function that characterizes a threshold policy. MA24 24-Room 401, Marriott Social Media Analytics and Big Network Data Sponsor: Artificial Intelligence Sponsored Session Chair: Bin Zhang, Assistant Professor, University of Arizona, Department of MIS, Tucson, AZ, 85721, United States of America, binzhang@arizona.edu 1 - Predicting Hacker IRC Participation using Discrete-time Duration Modeling with Repeated Events Anton Braverman, Cornell University, Ithaca, NY, 14850, United States of America, ab2329@cornell.edu, J. G. Dai

3 - Optimal Screening Policies for Women at High Risk of Breast Cancer

Caglar Caglayan, Georgia Institute of Technology, Atlanta, GA, United States of America, ccaglayan6@gatech.edu, Turgay Ayer, George Rust Women with breast density, family history of breast or ovarian cancer, or BRCA1 and BRCA2-mutation-carriers are at higher risk of breast cancer. For such women, non-mammographic modalities such as ultrasound or MRI, adjunct to or instead of mammogram, can be beneficial but they lead to an increased screening cost. Considering both potential health benefits and financial aspects, we study this multi-modality breast cancer screening problem and identify cost-effective optimal screening policies. 4 - Modeling Supply Chain System Structure to Trace Sources of Food Contamination Abigail Horn, PhD Candidate, Engineering Systems Division, Massachusetts Institute of Technology, 77 Massachusetts Ave., E40-240, Cambridge, MA, 02139, United States of America, abbyhorn@mit.edu, Richard Larson, Stan Finkelstein We are developing a methodology to identify high probability sources of contamination in the event of large-scale outbreaks of foodborne disease based on a graph-theoretic Bayesian inference algorithm. We present results from 2 modeling frameworks used to develop the inference algorithm: analytical models of stylized versions of the problem leading to new, general insights, and a Bayesian Network model used to support decision making and targeted information gathering during investigations. Sponsored Session Chair: Guodong Pang Assistant Professor, Penn State University, 363 Leonhard Bldg, University Park, PA, 16802, United States of America, gup3@engr.psu.edu 1 - Networks with Several Types of Interacting Tasks Yuri Suhov, Professor, University of Cambridge/Penn State University, yms@maths.cam.ac.uk I am going to present and discuss new analytic results on a class of reversible networks with several types of interactive customers. The main result is product- type formulas for the equilibrium distributions. 2 - Staffing Large-scale Service Systems with Stochastic Arrival Rates Ying Chen, PhD Candidate, University of Texas at Austin, 1 University Station, ETC 5.112, Austin, TX, 78712, United States of America, lesleycy@utexas.edu, John Hasenbein We minimize the staffing level for large-scale queueing systems with random arrival rates, where a QoS constraint is enforced on the probability a customer is queued. For single-station systems, we consider an Erlang-C model with only partial information provided for the discrete arrival-rate distribution, and we present a numerically stable procedure to obtain asymptotically optimal results. In the multi-station case, we introduce a joint QoS constraint and explore the corresponding solutions. 3 - Ergodic Control of Multiclass Multi-pool Networks in the Halfin- Whitt Regime Ari Arapostathis, Professor, University of Texas at Austin, Electrical and Computer Engineering, Austin, United States of America, ari@ece.utexas.edu, Guodong Pang We study the scheduling and routing control of Markovian multiclass multi-pool networks under the long-run average (ergodic) cost criteria in the Halfin-Whitt regime. We develop a new framework to study the associated ergodic diffusion controls and characterize the optimal solutions via the HJB equations. The asymptotic optimality results are established via a spatial truncation technique to approximate the solutions to the HJB. 4 - Pricing Server Information in Distributed Systems MA22 22-Franklin 12, Marriott New Advances in Stochastic Networks Sponsor: Applied Probability

Mauro Escobar, Columbia University, 500 W 120th Street, 3rd Floor, New York, NY, 10027, United States of America, me2533@columbia.edu, Mariana Olvera-Cravioto

We consider a queueing network where each job consists of a random number of pieces to be served in parallel, and such that all the pieces of a same job must begin their processing simultaneously. We analyze the performance of two models, one where the different pieces are routed to a random subset of servers and another one where they are optimally assigned to the servers with the shortest workloads. We illustrate how to use these two models to evaluate intermediate routing policies.

Victor Benjamin, University of Arizona, 1130 E Helen St, Room 430, Tucson, AZ, 85719, United States of America, vabenji@email.arizona.edu, Bin Zhang, Hsinchun Chen

Literature documents the existence of many active online hacker communities containing thousands of users. Some participants are expert cybercriminals, but

152