Low storage cost, partition-tolerant dynamic algorithms for replicated file systems

Abstract

Two algorithms are given for maintaining the consistency of replicated files. One is called dynamic supporting; the other is an extension of dynamic supporting called dynamic supporting with a greatest copy. The algorithms are hybrids of some existing algorithms. The correctness of these two algorithms is proved. Since replicas and votes are conceptually separated, the algorithms can achieve very good performance while still keeping storage costs very low. The proposed algorithms make no further assumptions about the distributed environment than conventional voting algorithms, so they can even tolerate network partition failures.<>