Temporal resources are defined as human or capital resources with a per-unit-time capacity that can be allocated to different services or products in different periods of time. Examples of temporal resources include machinery, computing power, warehouses, venues, staff, and specialized technology such as a chemical reactor. In this dissertation, I study the problem of dynamically allocating temporal resources to maximize revenue or to minimize costs when the decision-maker is uncertain about the outcome of decisions. I consider two different problems that represent challenges encountered in various industries. In the first chapter, I provide an introduction to the two problems presented in Chapters 2 and 3, discuss the respective motivating industries, and provide examples of broader applications.;The first problem, presented in Chapter 2, is the sales of cloud services to owners of interactive (user-based) applications such as websites and mobile apps. If an application owner purchases the service, the provider hosts the application on the cloud and provides the computing power required to support the application users. Here, the units of resource (hardware capacity) allocated to an application over time is directly determined by the traffic-pattern of the application's users. Considering the resource capacity, the provider dynamically prices services to maximize revenue. I model the provider's pricing problem as a large-scale stochastic dynamic program. I decompose this multi-dimensional stochastic dynamic program into single-dimensional sub-problems by proposing a tractable decomposition procedure. I then extend the proposed framework to define an individualized dynamic pricing mechanism for the cloud provider. To evaluate the performance of the proposed pricing mechanism, I present novel upper bounds on the optimal revenue. The computational results show that the proposed model of selling cloud services achieves significantly greater revenue than the prevalent alternative, and that the presented pricing scheme attains near-optimal revenue.;In the third chapter of my dissertation I analyze a catalyst-activated batch-production process with uncertainty in production times, learning about catalyst-productivity characteristics, and decay of catalyst performance across batches. The challenge is to determine the quality level of batches and to decide when to replenish a catalyst so as to minimize average costs consisting of inventory holding, backlogging, and catalyst switching costs. The temporal resource in this problem is the common reactor shared across batches and multiple products. I formulate this problem as a Semi-Markov Decision Process (SMDP), and use structural properties of the SMDP to define an effective two-level heuristic which is easy to interpret and implement, and to establish a lower bound on the optimal average cost to evaluate the heuristic. Through application to data from a leading food processing company, I show that the proposed methodology, in addition to attaining near-optimal costs, outperforms current practice by an average of 22 % reduction in costs.
展开▼
