关键词: 属性图, SPARQL查询, Spark GraphX, 查询树

Abstract: RDF (resource description framework) has become more and more popular due to its flexible and universal graph-like data model, and the size of RDF data is increasing in a rapid speed as well. It becomes infeasible and difficult to store and process RDF data on a single machine, which raises the requirement for the distributed parallel approaches. There are many approaches have been put forward on parallel computation framework, e.g., Hadoop and Spark for dealing with RDF storage and query. The major problem of the Hadoop-based methods is that the     intermediate results have to be written back to the disk, thereby incurring a lot of I/O operations. This paper proposes a novel method called SQX (SPARQL query on GraphX) for evaluating SPARQL (simple protocol and RDF query language) queries on Spark GraphX. By treating RDF data as a large property graph, it develops a new method to generate optimal query plans for SPARQL queries and executes them in a graph-parallel way. The SQX method adopts “query tree matching + result filtering” approach. Firstly, it generates the query tree and constraints from the SPARQL query and then in each super step, multiple query edges in the tree can be matched in a parallel fashion, and the final result will be generated from the iteration result which satisfies the constraints. Extensive experiments verify that this method is efficient and scalable in processing SPARQL queries on large-scale RDF data.

Key words: property graph, SPARQL query, Spark GraphX, query tree