Large RDF representation framework for GPUs case study key-value storage and binary triple pattern

Abstract

In this paper, we present the experimental framework which operates the search of RDF data in GPUs. We explore the use of triple storages for query processing in GPUs. From the user query, the SPARQL query is generated by the Middleware Jena. Then, the corresponding triples are searched. The triple search space is large and impossible to load into GPU memory to perform the parallel search. Proper representations are studied to storage data in GPUs for an effective search. We compare the key-value storage in Cassandra and binary triple storage using HDT. The experiments are done using the dataset, DBpedia and Freebase dataset. It is found that HDT compression can compress the data over ten times and the HDT format can be used to store large triples passed through the GPUs. The paper discusses qualitative and quantitative results from both manners. The pros and cons of them can further be combined properly in the further.