Preventing attacks in peer-to-peer document preservation systems

Librarians have centuries of experience preserving works published on many types of media from bone to paper. Academic publishing is moving to a digital process, thus shifting the problem of information preservation to that of preserving bits on digital storage media. Preserved digital documents may be damaged over long periods of time by storage system failures, operator errors, and malicious behavior. This thesis presents the LOCKSS (Lots of Copies Keeps Stuff Safe) digital preservation system that enables libraries to preserve their digital collections for very long time horizons. The LOCKSS system automatically collects journals from publishers' websites and preserves them through a peer-to-peer auditing protocol that repairs damage to digital documents. Because the LOCKSS system relies on the public Internet for communication between peer libraries, it is as susceptible as other distributed systems to network and protocol attacks. This thesis develops techniques to resist attacks against the LOCKSS system, presents an analysis of how the LOCKSS protocol behaves under a variety of attacks, and provides a methodology by which realistic adversary models can be designed and simulated.