Skip to Main content Skip to Navigation

Cohérence de données répliquées partagées par un groupe de processus coopérant à distance

Georges Brun-Cottan 1
1 SOR - Distributed Object Systems
Inria Paris-Rocquencourt
Abstract : Cooperative applications must communicate in order to make progress, and are therefore inherently affected by communication latency. In wide-area settings this latency can increase to a point where traditional consistency management no longer provides sufficient reactivity -- in some cases making the application unusable. We address this problem by considering a cooperative application to be a collection of cooperating replicas, and then concentrate on the management of consistency within such a group of replicas.

This problem is significant for two reasons: it appears in any application domain where we have cooperation between human agents, and is a fundamental part of the structure of cooperation mechanisms. Our comprehensive study of "weak consistency" covers four domains: transactional systems, distributed shared memories, distributed shared objects, and platforms for group communication. Our contribution has three significant aspects:

We present a new model of execution that is free of causal dependencies. It is based on replicated histories, and is used to formalise the consistency criteria presented in this document.

We present a new model of sharing, cooperative replication -- a derivative of active replication that does not impose a unique global ordering on the execution of operations. Replicas are autonomous except when their local view of the global computation can no longer guarantee correct behaviour of the distributed application.
We present a few system principles with which we define a new component, the consistency manager. This component:
manages the cooperation between replicas and implements the most complex parts of consistency management: the control of distribution, replication and concurrency;
provides a consistency contract that imposes deterministic properties on distributed operations. These can be considered as the "correctness criteria" for the distributed application;
is bound dynamically to the application at execution time; and

is reusable.
We have implemented Core -- a complete development environment conforming to the above principles -- that runs on Unix, is documented, and freely available by FTP. Core provides the usual services for distributed applications (a name service, group communication and management of group membership, for example), as well as an extensible class library with which specific consistency managers can be developed. It also provides numerous classes with which application designers can implement new kinds of shared object. Two example applications have been built using Core: an extension to Emacs that permits collaborative editing, and a simple manager for shared resources.
Complete list of metadata
Contributor : Georges Brun-Cottan <>
Submitted on : Thursday, July 5, 2007 - 8:22:31 PM
Last modification on : Wednesday, December 9, 2020 - 3:12:05 PM
Long-term archiving on: : Thursday, April 8, 2010 - 6:40:29 PM



  • HAL Id : tel-00160422, version 1


Georges Brun-Cottan. Cohérence de données répliquées partagées par un groupe de processus coopérant à distance. Réseaux et télécommunications [cs.NI]. Université Pierre et Marie Curie - Paris VI, 1998. Français. ⟨tel-00160422⟩



Record views


Files downloads