ITR/CISE: LOCKSS

Stanford University, Stanford CA

Investigators

Abstract

A highly dependable information infrastructure requires systems that are massively replicated, slow to fail, and resistant to attack. The LOCKSS system uses novel techniques for increasing dependability that show promise of wider applicability and increasing importance in this area, including: Building a fault-tolerant system from a large number of replicas that take part in "opinion polls" rather than a small number of replicas that take part in elections. Limiting the speed at which the system operates in order to limit the speed at which it can fail. Building a system that resists the activities of the "bad guy" without requiring that secrets be kept, in part by using public communication and a reputation system. Experience with this and other peer-to-peer systems shows that the real challenges of making systems work only become apparent at a realistic operational scale and running on real networks. The LOCKSS system is now running on the Internet with 60 peers worldwide. The project will exploit this LOCKSS test-bed to investigate areas in which further work is needed.