SPARQL Query Optimization on Top of DHTs

Abstract

Abstract. We study the problem of SPARQL query optimization on top of distributed hash tables. Existing works on SPARQL query processing in such environments have never been implemented in a real system, or do not utilize any optimization techniques and thus exhibit poor performance. Our goal in this paper is to propose efficient and scalable algorithms for optimizing SPARQL basic graph pattern queries. We augment a known distributed query processing algorithm with query optimization strategies that improve performance in terms of query response time and bandwidth usage. We implement our techniques in the system Atlas and study their performance experimentally in a local cluster.

Scalable distributed indexing and query processing over Linked Data

“…Similar to our work, [52] recently proposed databaseoriented query optimization strategies for RDF query processing on top of DHTs in the context of the Atlas system [53]. While one of our main goals is the optimization of the number of messages required for processing a query, in that work the authors explicitly focus on reducing the required bandwidth.…”

“…In contrast, UniStore focuses on less reliable systems and proposes to make extensive use of parallel processing strategies. [52] discusses very interesting and important extensions for our system, while on the other hand the Atlas system can benefit from the different processing strategies we discuss. An evaluation in a large local cluster of powerful machines shows that the idea of efficient RDF query processing in DHT systems can scale to millions of triples.…”

Scalable SPARQL querying using path partitioning

“…A popular data partitioning algorithm for RDF data is hash partitioning [13], [14], [18]. This approach distributes RDF triples across different partitions by computing a hash key over the subject or the object of each triple.…”

“…A popular approach to partition RDF data is hash partitioning, which is adopted by a majority of the existing distributed RDF engines [13], [14], [18], [24]. This approach distributes RDF triples across different partitions by computing a hash key over either the subject or the object of each triple.…”

D2R2: Disk-Oriented Deductive Reasoning in a RISC-Style RDF Engine

“…The bound-is-easier selection function is commonly used in recursive query evaluation, where the atom with the largest number of constants is evaluated first, in the hope of returning the smallest intermediate relation [14,16]. One can find extensions of this selection function in the literature, such as in [9] for a Semantic Web setting with binary predicates. There, the position of the bound argument is considered, where atoms with a bound first argument (subject) are preferred over those with a bound second argument (object) for two arguments with the same number of bindings.…”