A SPARQL query processing system using map-phase-multi join for big data in clouds

Abstract

Big data refers to large datasets which are huge, complex and hard to be stored and analysed by traditional data processing tools. Linked data is one of the approaches to deal with big data which are stored and processed in TripleStore. For huge dataset, TripleStore requires more scalable techniques. 'MapReduce' programming model is the most representative of cloud technology. There are several approaches using MapReduce to serve SPARQL query but still exhibit unacceptable performance in complex queries. In this paper, we propose a map-phase-multi-join algorithm for processing SPARQL queries. Using multi-join, job initialisation time is reduced by avoiding iterative of MapReduce jobs. Furthermore, map-phase join can save bandwidth by preventing join-less data to be transferred among computing nodes. We also design a storage schema and a join-order rule which enhance the performance of our system. The evaluation results show that our system outperforms traditional join approaches in most queries.