WaterFowl: A Compact, Self-indexed and Inference-Enabled Immutable RDF Store

Abstract

In this paper, we present a novel approach -called Water-Fowl -for the storage of RDF triples that addresses some key issues in the contexts of big data and the Semantic Web. The architecture of our prototype, largely based on the use of succinct data structures, enables the representation of triples in a self-indexed, compact manner without requiring decompression at query answering time. Moreover, it is adapted to efficiently support RDF and RDFS entailment regimes thanks to an optimized encoding of ontology concepts and properties that does not require a complete inference materialization or extensive query rewriting algorithms. This approach implies to make a distinction between the terminological and the assertional components of the knowledge base early in the process of data preparation, i.e., preprocessing the data before storing it in our structures. The paper describes the complete architecture of this system and presents some preliminary results obtained from evaluations conducted on our first prototype. 1

A Compact RDF Store Using Suffix Arrays

“…Now, we binary search the range Ψ [10,12] for the range that maps to 231 and find that Ψ [11] = 231. Then the final interval is R spo = [11,11].…”

“…The recent WaterFowl [12] was not included. Yet, since it reports space 10 times smaller than RDF-3X and similar times [12], we expect it would obtain worse query times than K2Triples (see comparison with RDF-3X below) and similar space. Other systems do not run over a dataset of this size [4].…”

“…The increasing interest in the management of RDF repositories (also called RDF stores) is witnessed by the various storage schemes proposed in recent years, which go from those based on relational databases [25] to native solutions such as BITMAT [6], RDF-3X [22], HEXASTORE [26], MonetDB [2], or WaterFowl [12]. As the RDF repositories grow in size, scalability issues challenge the use of RDF storage schemes [18].…”

Serializing RDF in Compressed Space

“…These subsets are highly compressed as (sparse) binary matrices that also allow efficient data retrieval. Other approaches, like HDT-FoQ [10] or WaterFowl [3] also enable data retrieval in compressed space. Both techniques, based on HDT serialization, report competitive performance at the price of using more space than k 2 -triples, which is the most effective compressor, to the best of our knowledge.…”

“…The previous reorganization ensures that all subjects within a given family have the same structure of predicates, as this was the main philosophy of grouping by predicate families. For instance, the example in Figure 5 shows that the first family comprises four predicates (1,3,5,7), so the three subjects in the family are related to these four predicates. This implies that each subject is related to four 1-bits in the object bitsequence 4 .…”