SARROD: SPARQL Analyzer and Reordering for Runtime Optimization on Big Data

Abstract

Resource Description Framework has been widely adopted for representing web resources and structured graph data model. Evolution of Big Data poses challenges in processing these graphs in terms of scalability as the size of the graphs may become enormously big. Tremendous growth of the data size (and in turn the graph size) will have an impounding effect on the execution times of the queries on the graph. Although big data platforms such as Hadoop can mitigate the problem, the query ordering and data flow between the constraints presents opportunities for further optimization. SPARQL, a widely used RDF query language suffers from the similar bottleneck for large graphs. There is hardly any established method to generate all equivalent reordering for a SPARQL query containing joins, outer joins, and group by aggregations. In this paper, we propose a query reordering algorithm viz., SARROD that leverages the property of the graphs that are simple to compute yet powerful for run time optimization. Experimental results show that SARROD reduces response time for SPARQL queries when executed over SHARD graph-store (triple-store) built on the Hadoop implementation of MapReduce by an order of 12% compared to non-ordered sequence.