An o(√N) algorithm for concurrency control in replicated file systems

Abstract

The problem of ensuring mutual exclusion in a distributed, replicated file system is investigated. An O(√N) algorithm is presented to solve the above problem and its correctness is proved. The nodes in the system are partitioned into mutually disjoint sets. By partitioning nodes appropriately, it is shown that a write quorum can be formed with 2√N - 1 copies, while a read quorum requires only √N copies. Comparisons are presented relating this algorithm to previous work. Mechanisms for handling the failures of links and sites are also proposed.