MuSe: a multi-level storage scheme for big RDF data using MapReduce

Abstract

Abstract Resource Description Framework (RDF) model owing to its flexible structure is increasingly being used to represent Linked data. The rise in amount of Linked data and Knowledge graphs has resulted in an increase in the volume of RDF data. RDF is used to model metadata especially for social media domains where the data is linked. With the plethora of RDF data sources available on the Web, scalable RDF data management becomes a tedious task. In this paper, we present MuSe—an efficient distributed RDF storage scheme for storing and querying RDF data with Hadoop MapReduce. In MuSe, the Big RDF data is stored at two levels for answering the common triple patterns in SPARQL queries. MuSe considers the type of frequently occuring triple patterns and optimizes RDF storage to answer such triple patterns in minimum time. It accesses only the tables that are sufficient for answering a triple pattern instead of scanning the whole RDF dataset. The extensive experiments on two synthetic RDF datasets i.e. LUBM and WatDiv, show that MuSe outperforms the compared state-of-the art frameworks in terms of query execution time and scalability.

Keywords

• RDF
• SPARQL
• Hadoop
• HDFS
• MapReduce
• Storage