Accès polyglotte aux systèmes de stockage multi stores de données massives : application aux données de mobilité et des smart-cities

Capture d’écran 2026-07-15 à 00.30.34

Soutenue par Léa El Ahdab, à l’Université de Toulouse, laboratoire IRIT le 21/11/2024. Co-Encadrée par Olivier Teste, André Péninou. Cadre magnifique avec un bel accueil. Sujet très proche de nos travaux au sein de l’équipe sur la gestion de bases de données distribuées. Très beaux échanges avec Olivier. [theses.fr] [examinateur]

This thesis scope is about querying with transparency distributed data in polystores. They are storage systems with flexibility which combine multiple SQL and NoSQL databases. This last family got popular with the proposition of a system allowing flexibility in the attributes data structures form. This advantage is linked to a disadvantage linked to data accessibility. With the context of data distribution, data can be vertically distributed where the structure is distributed in the different databases of the polystore; or horizontally distributed where instances are distributed in the databases of the polystores respecting a common structure. In practicality, hybrid distribution combining horizontal and and vertical distribution is a situation that complexifies data access for users. This brings the following issue: how users can access all distributed data stored in polystores environment containing different storage family systems? The work developed in this thesis was supported by the French Government in the framework of the Territoire d’Innovation program, an action of the Grand Plan d’Inversissement backed by France 2030, Toulouse M’etropole and the GIS neOCampus. They address the question of data querying with the proposition of unified models, which simplifies the polystore vision to the user, and an engine for query rewriting to adapt the user query to the different storage systems of the polystore. We distinct two types of distribution (vertical and horizontal) despite similarities in the solution development. This thesis proposes a framework divided in two part. The first one build unified models and equivalences of attributes structures between their position in the polystore and the one in the unified models. These information are stored in a mapping dictionary. It helps the rewriting of the user query to access to distributed data in the polystore. It is updated when the data insert changes the attribute structure in the polystore. The second part of the framework is the exploitation of these mapping dictionaries and rewrite the user query (initially expressed in one language according to the chosen unified model) accordingly to the real position of data inside the polystore. We have studied existing works responding to the same issue and we have classified them by their solution approach. This thesis is compared to these works with multiple criteria (storage systems in the polystore, querying language, vision of the polystore, data distribution, etc.) in order to position our solution. We have defined key concepts to understand the scope of this thesis: polystore, database, dataset, attribute, operator, mapping, structural heterogeneity, data fragmentation, fragmentation key. Notions of translation from a storage system to another and from a specific algebra into its associated language are also developed in the query rewriting process. It produces a query more complete from the initial user query and keeps DBMS operators it interrogates. We show how it is possible to solve the issue of data distribution complexity in a polystore with unified models and query rewriting. Finally, we have experimented our solution with several scenarios to expend tests and results. We evaluate the query rewriting time and the mapping dictionaries generation according to data distribution, data volumes, heterogeneity rate in the NoSQL document-oriented system; and the query execution time depending on data distribution, data volumes and heterogeneity rate. Our experiments were conducted on two different and independent datasets (TPC-H, Unibench). Then, we proposed our solution in Uruguay to allow the research team of the University of Montevideo to expend their scope to calculate their data quality indicators. We have studied data mobility in order to test our framework in the context of neOCampus project from the data collected on the campus of the University Toulouse III, Paul Sabatier.

Laisser un commentaire