HIDALGO CASTILLO Nicolas

PhD graduated
Team : REGAL
Departure date : 12/01/2011
https://lip6.fr/Nicolas.Hidalgo

Supervision : Pierre SENS

Co-supervision : ARANTES Luciana

Towards an Efficient Support for Complex Queries on Structured Peer-to-Peer Networks

Distributed Hash Tables (DHTs) provide the substrate to build scalable, structured, and efficient Peer-to-Peer (P2P) networks which are distributed systems with the potential to handle massive amounts of data on a very large scale. Due to the hashing assignment of object keys to peers and its lookup operation, exact match queries present very good performance on DHT-based P2P systems. Although, traditional DHTs cannot provide an effective support for complex queries, such as range queries. In this work, we are particularly interested in those solutions based on prefix tree indexes since they provide a portable and scalable approach for satisfying complex queries over DHTs. Nevertheless, the search methods proposed by such solutions usually generate both high latencies and unnecessary message traffic overhead which degrade system performance. Furthermore, some of them present load balancing problems and do not tolerate P2P churn. In this thesis we have proposed two solutions: PORQUE and ECHO. The former is oriented to support low-latency searches while the latter is oriented to support low-overhead searches. Performance evaluation results of experiments confirm that both PORQUE and ECHO can reduce latency and message traffic of searches by more than 50 % when compared to PHT. Our solutions also offer load balancing minimising as much as possible bottlenecks over the index structure. By exploiting different datasets distributions, performance results show that our solutions perform independently of data skewness. Moreover, performance of PORQUE and ECHO do not degrade in dynamic environments.

Defence : 11/29/2011 - 16h - Site Jussieu 25-26/105

Jury members :

Anne-Marie Kermarrec, Directrice de Recherche INRIA - Rapporteur
Claudia Roncancio, Professeur Université de Grenoble - Rapporteur
Peter Druschel, Scientific Director Max Planck Institute for Software Systems
Franck Petit, Professeur Université Pierre et Marie Curie
Xavier Bonnaire, Professeur Associé Universidad Tecnica Federico Santa Maria
Pierre Sens, Professeur Université Pierre et Marie Curie
Luciana Arantes, Maitre de Conference Université Pierre et Marie Curie

2010-2016 Publications