Performance

Performance

High performance in standard reasoning tasks is the primary goal of ELK. This is achieved by using highly efficient consequence-based reasoning algorithms that have been further enhanced to take advantage of modern multi-core processors. However ELK offers very competitive performance even on single-core CPUs.

Experimental Results

Initial versions of ELK illustrated a very good performance on standard ontologies such as SNOMED CT which could be classified in under 5 sec on a laptop. For detailed performance results, please refer to the publications on the main page.

How to Measure ELK Performance

The best way for measuring ELK performance is to use the command line client which reports times for each major processing step (loading/parsing, saturation, transitive reduction for classification, etc.). Reasoning is mainly bounded by CPU speed, but sufficient memory is required to store all inferences. In particular, even if no out-of-memory error occurs, less memory leads to an increased garbage collection activity in Java that will take additional time.