Master thesis : Toward functional and distributed R2RML processor

Abstract

Resource Description Framework (RDF) offers multiple advantages for data storage. Transforming data from relational databases into RDF datasets can be interesting. One prominent approach for generating RDF datasets from relational databases is the W3C relational database to RDF (R2RML) mapping language. Existing R2RML processors face challenges related to computing time and memory consumption, particularly when dealing with large-scale relational databases. This master's thesis presents a functional and distributed solution for implementing an R2RML processor working on cluster. A Scala solution based on Apache Spark that is purely functional is proposed. This approach involves an updated Java Parser from an existing implementation, a transformation of Java objects into Scala Abstract Data Type (ADT), a preprocessing to rewrite referencing object map into new triples map, and the generation and writing of the data. In this solution, the distribution of the task is based on relational data rows. For modestly-sized databases, this solution is slow due to an overhead introduced by Apache Spark. While being computed on cluster, the solution is fast for generation and will not consume too much memory. But, on too large-scale data, it suffers from memory problems that can be solved.