mStore: Schema Mining based-RDF Data Storage

Guopeng Zheng, Tenglong Ren, Lulu Yang, Xiaowang Zhang, Zhiyong Feng

Abstract

The relationship-based approach is an efficient solution strategy for distributed RDF data management. The schema of tables can directly affect the system’s storage efficiency and query performance. Most current approaches are based on fixed schema(e.g., VP, ExtVP). When facing large-scale RDF datasets and complex SPARQL queries requiring many joins, such methods suffer from problems such as long pre-processing time and poor query performance. Schemas with Pareto Optimality between the system’s space consumption and query efficiency are needed but also hard to find. Therefore, we propose mStore, a prototype system with flexible schemas based on schema mining. The intuition behind our approach is that we want to divide the combinations of predicates with higher relevance into the same schema, which can replace costly joins with low-cost selects, improving the query performance. The results show that our system performs better on complex query workloads while reducing the pre-processing time overhead compared to systems with fixed schema partitioning strategies.
BibTeX
@inproceedings{zheng2023mstore,
  author = {Guopeng Zheng and Tenglong Ren and Lulu Yang and Xiaowang Zhang and Zhiyong Feng},
  title = {{mStore: Schema Mining based-RDF Data Storage}},
  booktitle = {The Web Conference},
  year = {2023},
  doi = {10.1145/3543873.3587339},
}

Loading related content...

Figure