Skip to Main Content
Peer-to-peer file systems are typically vulnerable to denial-of-service or free-loader problems. Those that address these issues employ approaches that are either simplistic or require a centralized authority. We explore how a zero-sum trading system can provide strong quotas to a peer-to-peer distributed file system without any centralized authority. We treat each member of such a peer-to-peer system as an autonomous agent, interested in preserving its own disk storage. We develop a model for these agents, and present experimental simulation and external emulation results from a multi-agent reinforcement learning model demonstrating the validity of this approach.
Date of Conference: 17-20 Dec. 2002