Skip to Main content Skip to Navigation

Gestion et optimisation des données massives issues du Web

Abstract : Big Data represents a challenge not only for the socio-economic world but also for scientific research. Indeed, as has been pointed out in several scientific articles and strategic reports, modern computer applications are facing new problems and issues that are mainly related to the storage and the exploitation of data generated by modern observation and simulation instruments. The management of such data represents a real bottleneck which has the effect of slowing down the exploitation of the various data collected not only in the framework of international scientific programs but also by companies, the latter relying increasingly on the analysis of large-scale data. Much of this data is published today on the WEB. Indeed, we are witnessing an evolution of the traditional web, designed basically to manage documents, to a web of data that allows to offer mechanisms for querying semantic information. Several data models have been proposed to represent this information on the Web. The most important is the Resource Description Framework (RDF) which provides a simple and abstract representation of knowledge for resources on the Web. Each semantic Web fact can be encoded with an RDF triple. In order to explore and query structured information expressed in RDF, several query languages have been proposed over the years. In 2008,SPARQL became the official W3C Recommendation language for querying RDF data.The need to efficiently manage and query RDF data has led to the development of new systems specifically designed to process this data format. These approaches can be categorized as centralized that rely on a single machine to manage RDF data and distributed that can combine multiple machines connected with a computer network. Some of these approaches are based on an existing data management system such as Virtuoso and Jena, others relies on an approach specifically designed for the management of RDF triples such as GRIN, RDF3X and gStore. With the evolution ofRDF datasets (e.g. DBPedia) and Sparql, most systems have become obsolete and/or inefficient. For example, no one of existing centralized system is able to manage 1 billion triples provided under the WatDiv benchmark. Distributed systems would allow under certain conditions to improve this point but consequently leads a performance degradation. In this Phd thesis, we propose the centralized system "RDF_QDAG" that allows to find a good compromise between scalability and performance. We propose to combine physical data fragmentation and data graph exploration."RDF_QDAG" supports multiple types of queries based not only on basic graph patterns but also that incorporate filters based on regular expressions and aggregation and sorting functions. "RDF_QDAG" relies on the Volcano execution model, which allows controlling the main memory, avoiding any overflow even if the hardware configuration is limited. To the best of our knowledge, "RDF_QDAG" is the only centralized system that good performance when manage several billion triples. We compared this system with other systems that represent the state of the art in RDF data management: a relational approach (Virtuoso), a graph-based approach (g-Store), an intensive indexing approach (RDF-3X) and two parallel approaches (CliqueSquare and g-Store-D). "RDF_QDAG" surpasses existing systems when it comes to ensuring both scalability and performance.
Document type :
Complete list of metadata
Contributor : Abes Star :  Contact
Submitted on : Thursday, December 17, 2020 - 9:41:09 AM
Last modification on : Friday, March 26, 2021 - 10:27:37 AM
Long-term archiving on: : Thursday, March 18, 2021 - 6:32:01 PM


Version validated by the jury (STAR)


  • HAL Id : tel-03079086, version 1



Abdallah Khelil. Gestion et optimisation des données massives issues du Web. Autre [cs.OH]. ISAE-ENSMA Ecole Nationale Supérieure de Mécanique et d'Aérotechique - Poitiers; Université d'Oran, 2020. Français. ⟨NNT : 2020ESMA0009⟩. ⟨tel-03079086⟩



Record views


Files downloads