Towards optimization of RDF analytical queries on MapReduce

摘要

The broadened use of Semantic Web technologies across domains has led to a shift in focus from simple pattern matching queries on RDF data to analytical queries with complex grouping and aggregations. An RDF analytical query involves graph pattern matching, which translates to several join operations due to the fine-grained nature of RDF data model. Complex analytical queries involve multiple grouping-aggregations on different graph patterns, making such tasks join-intensive. Scale-out processing of RDF analytical queries on existing relational-style MapReduce platforms such as Apache Hive and Pig, results in lengthy execution workflows with multiple cycles of I/O and network transfer. Additionally, certain graph patterns result in avoidable redundancy in intermediate results, which negatively impacts processing costs. The PhD thesis summarized in this paper proposes a two-pronged approach to minimize the costs while processing RDF queries on MapReduce: an algebraic approach based on a Nested TripleGroup Data Model and Algebra that reinterprets graph pattern queries in a way that reduces the required number of map-reduce cycles, and special strategies to minimize the redundancy in intermediate data while processing certain graph patterns. The proposed techniques are integrated into Apache Pig. Empirical evaluation of this work for processing graph pattern queries show 45–60% performance gains over systems such as Pig and Hive.