Atomicity vs. availability : concurrency control for replicated data

Abstract

Data managed by a distr ibuted program may be subject to consistency and availability requirements that must be satisfied in the presence of concurrency, site crashes, and network partit ions. This paper proposes two integrated methods for implementing concurrency control and replication for data of abstract type. Both methods use quorum consensus. The Consensus Locking method minimizes constraints on availability, and the Consensus Schedul ing method minimizes constraints on concurrency. These methods systematically exploit type-specif ic propert ies of the data to provide better availability and concurrency than methods based on the convent ional read/wr i te classif ication of operat ions. Necessary and suff icient constraints on correct implementations are derived directly from the data type specif icat ion. These constraints reveal that an object cannot be replicated in a way that simultaneously minimizes constraints on both availability and concurrency. C o p y r i g h t © 1985 Maurice Herlihy i' This research was sponsored by the Defense Advanced Research Projects Agency (DOD), ARPA Order No. 3597, monitored by the Air Force Avionics Laboratory Under Contract F33615-81-K-1539 The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of the Defense Advanced Research Projects Agency or the US Government.