Ownership and decentralization in distributed systems allocation mechanisms
Abstract
Resource allocation is one of the most widely addressed topics in computer science. Its importance lies with the role it has in providing a foundation for building systems and applications, and with its focus on using the resources available in the environment in an efficient way. While, generally, the problem and the goals of resource allocation remain the same---efficient allocations according to certain parameters---the environments in which it happens vary in many ways. We focus on rethinking the allocation problem in the setting of a decentralized environment, such as the Internet, where resources belong to independent and selfish participants. We are using mechanism design theory, a combination of economics and game theory, to research and design auctions as allocation mechanisms in such a way as to implement consensus and efficient outcomes when resource owners are selfish. While previous research in this area has focused on centralized auction mechanisms under various resource-or-participant scenarios, we are looking at complete environments as a whole. Our goal is to design mechanisms that produce an efficient allocation across the whole set of auctions in the market. Decentralized markets exhibit three characteristics, which this thesis addresses directly. The first one is the spatial fragmentation of the allocation mechanism, which happens due to the independence of resource owners running the auctions. In this case, we show that the market should support an indirect auction with historical and provisional information in order to obtain efficient allocations. The second one is the fragmentation of resource ownership, which happens due to users' needs to have several resources: algorithms and physical resources. For this situation, we propose a proxy-based resource acquisition, and we show how to divide the consumer's bid for the needed resources. The third characteristic of such a large-scale allocation with selfish participants is the, fragmentation of the trust associated with the interacting parties. Our solution offers a decentralized scheme for trust management, based on market behavior (e.g. on payment defaulting) that protects the allocation against the participants' misbehavior.
Degree
Ph.D.
Advisors
Rego, Purdue University.
Subject Area
Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.