Informs Annual Meeting 2017

WA19

INFORMS Houston – 2017

3 - Capacity Expansion Strategies in Cloud Computing Environment Mohammad Ebrahim Arbabian, PhD Student, University of Washington, 4120 Brooklyn Ave NE unit 503, # 4, Seattle, WA, 98105, United States, arbabian@uw.edu, Shi Chen, Kamran MoeinZadeh We study the problem of deploying new servers in a cloud company. We consider a continuous review system and present the optimal policy which minimizes the total discounted cost rate in finite and infinite horizon settings. Moreover, the optimal server specifications to be ordered each cycle given the current servers are derived. 4 - Managing Innovation Pipeline by Small Firm Researchers Mengyang Pan, Southwest University of Finance and Economics, Small firm researchers may have multiple R&D projects in their pipeline, that are in various development stages and may or may not have substantial external collaborations. This research intends to explore the challenges for them induced by limited time and resources, as well as IP protection risks. Our analyses of longitudinal data suggest the timing and extent of external collaboration in a portfolio of projects can be an important decision for success. 5 - Dynamic Task Allocation with Learning and Forgetting Thomas Vossen, University of Colorado-Boulder, Leeds School of Business, UCB419, Boulder, CO, 80309, United States, thomas.vossen@colorado.edu, Peter Letmathe We consider a setting where tasks arrive randomly over time for possible processing. Incoming tasks can be allocated to (human) resources, whose productivity depends on the number of tasks processed by the resource before (learning) and is impacted by changes in the workforce over time (forgetting). We formulate the task allocation problem as a weakly coupled stochastic dynamic programming problem, and use a Lagrangian Relaxation approach to derive heuristic allocation policies. We evaluate the flexibility and resilience that emerge from these policies, and analyze how various environmental factors impact performance. 6 - Linking Operations Characteristics to Operations Strategy in Hospital Merger and Acquisitions Yuqiao Cheng, University of Houston, 4800 Calhoun Rd, Houston, TX, 77004, United States, chengyuqiao126@gmail.com, Xiaosong David Peng We present research examining the impact of hospital M&A at the operational level. The impact may stem from similarities in operational characteristics and a more focused operations strategy. We analyze archival data from various sources using Difference in Differences analysis and conduct case studies to present our findings. 692 Riverview Drive, Apt 128, Columbus, OH, 43202, United States, pan.295@osu.edu, Aravind Chandrasekaran 340A Big Data and Queueing Networks Sponsored: Applied Probability Sponsored Session Chair: Cathy Honghui Xia, Ohio State University, Ohio State University, Columbus, OH, 43210, United States, xia.52@osu.edu 1 - Queueing-theoretic Models for Data Center Performance Mor Harchol-Balter, Carnegie Mellon University, 5000 Forbes Avenue, Computer Science Dept, Pittsburgh, PA, 15213, United States, harchol@cs.cmu.edu Data Centers are a never-ending source of interesting performance problems for queueing theorists. In this talk we will introduce problems ranging from Job Replication to Hadoop Scheduling to Optimizing Parallelism. For each problem, we will propose queueing-theoretic models (and some solutions), with an emphasis on realistic assumptions. This talk is meant to bridge Computer Systems optimization and OR modeling/analysis. 2 - Queueing Control using Learning Based Semi-online Algorithms WA17

3 - Throughput Scalability of Large Parallel and Distributed Processing Systems with Advancing Capabilities Yun Zeng, The Ohio State University, Columbus, OH, United States, zeng.153@buckeyemail.osu.edu, Cathy Honghui Xia Parallel and distributed processing systems have expanded in great scale in cloud computing and big data analytics. A critical issue concerns the throughput scalability: whether or not the throughput sustains as the systems scale in size. This issue is highly non-trivial as the systems can have complicated structures and advancing capabilities. We present necessary and sufficient conditions on throughput scalability of parallel and distributed processing systems with arbitrary structure and generally distributed service times. In particular, we present conditions under which scaling storage and scaling processing speed can recover throughput even when the system is not well-structured. 342A Procurement and Purchasing Management Contributed Session Chair: Wei-Shiun Chang, National Cheng Kung University, Tainan City 701, Taiwan, wschang@mail.ncku.edu.tw 1 - Assortment Decisions with Substitutability and Complementarity Xi Shan, University of Texas at Dallas, Richardson, TX, 75080, United States, xxs130630@utdallas.edu, Dorothee Honhon, Suresh P.Sethi We consider the assortment plan of a retailer with complementary or substitutable goods/categories. We find that increasing the variety in one category can stimulate more variety in the other complementary category. For a single category retailer, increasing variety for that category will attract more customers but cost more on inventory. Now for a retailer with complementary categories, increasing variety for one category will not only bring more attraction and inventory cost, but also generate more customers having the willingness to visit the other category. And the other category can take this chance to earn more profit. 2 - Dynamic Optimization Models for Sourcing Problems Nilofar Varzgani, PhD Candidate, Rutgers Business School, 113 Harrison Avenue, Apt 204, Harrison, NJ, 07029, United States, nilofarv@scarletmail.rutgers.edu, Michael N.Katehakis When manufacturers are sourcing components there always comes the dilemma of deciding between having 1 supplier and using volume leverage to get the best possible cost advantage. On the other hand, this poses a supply disruption risk . Therefore, most manufacturers opt for the multi-supplier model where they have dual or more procurement requirements. The decision problem faced by the manufacturing company is twofold; election problem and volume allocation problem. We solve for this optimal selection and allocation problem as a one step process using a dynamic programming framework in a contract manufacturing framework. 3 - OEM Selling Channels and Supply Chain Performance This paper studies how an OEM’s selling channel can serve as an instrument to enhance its product accessibility in the local market for developing countries. we find that authorizing the OEM selling channel can be a win-win strategy for both the brand (gaining more market share in the international market) and the OEM (earning more profits) because of economy of scales that lowers unit production costs. 4 - Renegotiation in Procurement Auctions with Endogenous Liability Wei-Shiun Chang, National Cheng Kung University, No.1, University Road, Tainan City 701,, 701, Taiwan, wschang@mail.ncku.edu.tw, Rhea Salonga The winners in the procurement auctions may find the auction prices unfavorable due to the Winner’s Curse phenomenon. Practically buyers offer an opportunity to winning sellers to renegotiate the price in such a case in the hope of preventing sellers from bankrupt. We derive that sellers would be better off to delegate a subsidiary with less liability to bid on behalf of parent companies when renegotiation is allowed at the expense of procurers. We then test the theoretical predictions with experiments. The findings have important implications for procurement policy reform as well as to the procurers’ selection of mechanisms (i.e. whether to use renegotiation) to resolve potential bankruptcy issues. Chia-Wei Kuo, National Taiwan University, 85 Sec 4, Roosevelt Rd, Taipei, 106, Taiwan, cwkuo@ntu.edu.tw WA19

Jin Xu, Texas A&M.University, 3120 Texas A&M.University, College Station, TX, 77840, United States, jinxu@tamu.edu, Natarajan Gautam

Motivated by applications in smart reconfigurable manufacturing, we consider queueing systems where workload is revealed before service with the help of IoT. The objective is to minimize sojourn time for jobs under constraints of order of service and server conditions. We propose a learning-based semi-online algorithm to solve the problem and demonstrate its performance.

447