Abstract
In real time collaborative systems, replicated objects, shared by users, are subject to concurrency constraints. In order to satisfy these, various algorithms, have been proposed that exploit the semantic properties of operations and use Operational transformations to serialize concurrent operations and achieve copy convergence of replicated objects, while preserving users intention. Their drawback is that they either require a condition on user's operations which is hard to verify when possible to ensure, or they need undoing then redoing operations in some situations. The fi rst objective of the thesis is to present two new algorithms that overcome these drawbacks. They are based upon the implementation of a continuous global order which enables that condition to be released, and simpli es the operation integration process. In one of these algorithms, SOCT4 for which we give a formal proof, this process becomes even more simplified thanks to deferred broadcast whereas in the other one, SOCT5, priority is given to concurrency during the broadcast. An extension to these algorithm to support mobile sites is introduced. The second objective is to adapt SOCT2 type algorithms to make it possible for a user to undo an operation insofar as the rare nonrestrictive proposals made to solve this problem compromise in certain cases the convergence of the copies. For that, rather than to handle the opposite operation directly, we consider undo as a speci c operation whose forward transposition must satisfy two general conditions that the study highlights. The reasonableness of this process is validated by the study of critical cases.