Méthodologie d'évaluation pour les types de données répliqués

Mehdi Ahmed-Nacer 1
1 SCORE - Services and Cooperation
Inria Nancy - Grand Est, LORIA - NSS - Department of Networks, Systems and Services
Abstract : To provide a high availability from any where, at any time, with low latency, data is optimistically replicated. This model allows any replica to apply updates locally, while the operations are later sent to all the others. In this way, all replicas eventually apply all updates, possibly even in different order. Optimistic replication algorithms are responsible for managing the concurrent modifications and ensure the consistency of the shared object. In this thesis, we present an evaluation methodology for optimistic replication algorithms. The context of our study is collaborative editing. We designed a tool that imple- ments our methodology. This tool integrates a mechanism to generate a corpus and a simulator to simulate sessions of collaborative editing. Through this tool, we made several experiments on two different corpus : synchronous and asynchronous. In synchronous collaboration, we evaluate the performance of optimistic replication algorithms following several criteria such as execution time, memory occupation, message’s size, etc. After analysis, some improvements were proposed. In addition, in asynchronous collaboration, when replicas synchronize their modifications, more conflicts can appear in the document. In this case, the system cannot merge the modifications until a user resolves them. In order to reduce the conflicts and the user’s effort, we propose an evaluation metric and we evaluate the different algorithms on this metric. Afterward, we analyze the quality of the merge to understand the behavior of the users and the collaboration cases that create conflicts. Then, we propose algorithms for resolving the most important conflicts, therefore reducing the user’s effort. Finally, we propose a new architecture for supporting cloud- based collaborative editing system. This architecture is based on two optimistic replication algorithms. Unlike current architectures, the proposed one removes the problems of the centralization and consensus between data centers, is simple and accessible for any developers.
Complete list of metadatas

Cited literature [108 references]  Display  Hide  Download

Contributor : Mehdi Ahmed-Nacer <>
Submitted on : Thursday, January 7, 2016 - 1:14:17 PM
Last modification on : Tuesday, December 18, 2018 - 4:26:02 PM
Long-term archiving on: Friday, April 8, 2016 - 1:22:46 PM


  • HAL Id : tel-01751533, version 2


Mehdi Ahmed-Nacer. Méthodologie d'évaluation pour les types de données répliqués. Calcul parallèle, distribué et partagé [cs.DC]. Université de Lorraine, 2015. Français. ⟨NNT : 2015LORR0039⟩. ⟨tel-01751533v2⟩



Record views


Files downloads