This award provides funding for the development of a mathematical theory of large-scale service systems that employ flexible resources in order to process a sequence of diverse incoming tasks. Centralized systems with fully flexible resources (so-called complete resource pooling) generally offer significant performance improvements when compared to decentralized and inflexible systems in which each resource is dedicated to a specific type of tasks. However, full flexibility typically entails a heavy infrastructural cost. The project will focus on identifying and analyzing situations where a limited amount of flexibility (i.e., when only a small fraction of the available resources are flexible) achieves almost all of the performance benefits of full flexibility. In particular, the project will study various mathematical models involving partial flexibility and/or partial information sharing, and carry out a performance analysis, with a focus on systems that involve nontrivial dynamics, and in the limit of large systems, in order to derive scaling laws.
Partial flexibility has the potential for major performance improvements, at moderate capital and operating cost, in a vast array of important application domains, such as data centers, server farms, sensor networks, supply chains, and medical staffing. If successful, the results of this research (to be disseminated through scholarly publications, seminars, and the training of students) will provide valuable insights and guidance on how to architect, design, and operate efficiently partially flexible systems.