A Comparative Study of Stream Reasoning Engines

The diverse research efforts in recent years in the area of stream reasoning (SR) led to a wide range of SR engines. However, the lack of standardization and the diverse choices in SR (e.g., tuple-driven vs. time-driven engines, streaming all results vs. newly derived ones, ...) mean that real comparability among the engines is hardly given. A first step towards achieving comparability and standardization is the RSP-QL model, implemented in the RSP4J framework, which allows for describing and formalizing the semantics of SR engines. To further advance the state of the art in comparative research of stream reasoning, we present the results of a survey to quantify the in-use importance of several key performance indicators (KPIs) and features and compare SR engines along these KPIs with the CityBench and the CSRBench oracle. Our analysis shows that the two RSP4J implementations C-SPARQL2.0 and YASPER outperform the well-known C-SPARQL implementation in terms of performance and configurability. Our comparison against a naive SR extension of the incremental reasoning engine RDFox shows that SR engines still have potential for improvement. To avoid a costly integration of engines into several different benchmarking environments, we finally present a unifying interface, already aligned with the CityBench and CSRBench, for benchmarking SR engines.