Massive-scale RDF Processing Using Compressed Bitmap Indexes

Conference · Thu May 26 04:00:00 EDT 2011

OSTI ID:1056553

Madduri, Kamesh; Wu, Kesheng

The Resource Description Framework (RDF) is a popular data model for representing linked data sets arising from the web, as well as large scienti c data repositories such as UniProt. RDF data intrinsically represents a labeled and directed multi-graph. SPARQL is a query language for RDF that expresses subgraph pattern- nding queries on this implicit multigraph in a SQL- like syntax. SPARQL queries generate complex intermediate join queries; to compute these joins e ciently, we propose a new strategy based on bitmap indexes. We store the RDF data in column-oriented structures as compressed bitmaps along with two dictionaries. This paper makes three new contributions. (i) We present an e cient parallel strategy for parsing the raw RDF data, building dictionaries of unique entities, and creating compressed bitmap indexes of the data. (ii) We utilize the constructed bitmap indexes to e ciently answer SPARQL queries, simplifying the join evaluations. (iii) To quantify the performance impact of using bitmap indexes, we compare our approach to the state-of-the-art triple-store RDF-3X. We nd that our bitmap index-based approach to answering queries is up to an order of magnitude faster for a variety of SPARQL queries, on gigascale RDF data sets.

• Research Organization: Ernest Orlando Lawrence Berkeley National Laboratory, Berkeley, CA (US)
• Sponsoring Organization: Computational Research Division
• DOE Contract Number: AC02-05CH11231
• OSTI ID: 1056553
• Report Number(s): LBNL-5316E
• Country of Publication: United States
• Language: English

Related Subjects

97 MATHEMATICS AND COMPUTING
RDF
SPARQL query optimization
compressed bitmap indexes
large-scale data analysis
semantic data