Identifying and determining SPARQL endpoint characteristics. Issue 3 (19th August 2014)

• Record Type: Journal Article
• Authors: Lorey, Johannes
• Editors: Taniar, David; Pardede, Eric
• Is Part Of: International journal of web information systems. Volume 10:Issue 3(2014)
• Publication Date: 2014-08-19
• Subjects: World Wide Web -- Periodicals; Internet -- Periodicals; Information storage and retrieval systems -- Periodicals; 004.678
• DOI: 10.1108/IJWIS-03-2014-0007
• Languages: English
• ISSNs: 1744-0084

Abstract

Abstract : Purpose: Publicly accessible SPARQL endpoints contain vast amounts of knowledge from a large variety of domains. However, oftentimes these endpoints are not configured to process specific workloads as efficiently as possible. Assisting users in leveraging SPARQL endpoints requires insight into functional and non-functional properties of these knowledge bases. In this work, we introduce several metrics that enable universal and fine-grained characterization of arbitrary Linked Data repositories. Design/methodology/approach: We present comprehensive approaches for deriving these metrics. More specifically, we utilize concrete SPARQL queries to determine corresponding values. Furthermore, we validate and discuss the introduced metrics through extensive evaluation on real-world SPARQL endpoints. Findings: We determined in our evaluation that endpoints exhibit different characteristics: While it comes as no surprise that latency and throughput are influenced by the network infrastructure, the costs for join operations depend on a number of factors that are not obvious to a data consumer. Moreover, as we discuss mean, median, and upper quartile values, we found both endpoints behaving consistently as well as repositories offering varying levels of performance. Originality/value: On the one hand, the contribution of our work lies in assisting data consumers in their evaluation of the quality of service of publicly available SPARQL endpoints. On the other hand, the performance metrics introduced in this paper can also be considered as additional input features for distributed query processing frameworks. Moreover, we provide a universal means for discerning characteristics of different SPARQL endpoints without the need of (synthetic or real-world) query workloads.