On fault-tolerant data replication in distributed systems

This paper presents a new fault-tolerant algorithm for dynamic data replication in distributed systems. The algorithm presents remedies to the deficiencies of the existing adaptive data replication (ADR) and the primary missing writes (PMW) algorithms, proposed in [ACM Trans. Database Syst. 22 (1997) 255]. The new algorithm extends the idea of PMW to allow the adaptive replication algorithm to be used in the presence of failures, and introduces new mechanisms for maximising object availability. Results from a comparative analysis reveal that our algorithm outperforms the ADR/PMW especially in a read intensive environment like the Internet.