| Abstract |
Poor performance, and non-predictability have long been the challenges of reasoning on RDF data. Therefore, vast studies have been conducted to mitigate these problems by improving the reasoning algorithms. Most existing efforts though useful are incomplete, because 1) they attempt to address the problem for the worst-case scenario and in very high level, 2) they do not consider the effect of underlying computer system on behavior of reasoner as a computer program. Given the fact that the worst case happens very infrequently and the overhead of underlying computer system is not negligible for programs with special demands like a reasoner, we believe there is still a great chance of improvement for RDF reasoners. To that end, we advocate a different approach which devises adaptations to reasoners for better performance and more predictability based on observations of reasoning on real-world data. Our goal is to leverage the characteristics of input data as well as spotting and resolving the system bottlenecks to boost the performance and predictability of RDF reasoners. |