During the last decade, a computing paradigm known as “grid” computing has seen a surge in research activity. Drawing inspiration from the electrical power grid, this paradigm is an approach to high-performance computing that seeks to serve a large number of users from multiple, geographically distinct computing centers. As grids exist today, users are competing for overcommitted resources. Without a well-designed mechanism to mediate the competing interests of users, the outcome can be chaotic and inefficient. Additionally, scheduling decisions are made based on user-submitted metadata, data that can be manipulated to increase a user’s share of resources.

This dissertation explores the efficacy of auction-based schedulers as a means for mediating these competing interests. In particular, the use of auctions to incentivize honest disclosure of job metadata is investigated. We investigate this problem in the context of best-effort batch queues and reservation systems.