Hi OpenNLP models are referenced with WeakReferences. So if they are not used for some time the GC might decide to garbage collect them. In this case they well need to be loaded again fro the next request. However I would expect this to take in the range of 500ms to 2sec and not longer. Similar things are true for other integrated NLP frameworks. E.g. when using Stanbol-Freeling integration via the RESTful NLP processing engine you might also experience an increase of 2sec up to 10sec when Freeling needs to load its NLP processing models.
Entityhub Sites - such as dbpedia - do need no initialization times. They are active permanently. If a Site is not available the EnhancementEngine will simple fail with an according exception. To Summarize: Thee might be cases where a request can take some additional seconds, but AFAIK there is no reason why a request of a short document should take longer than that. If you want details about the execution times of the different engines you can add the query parameter "executionmetadata=true" to enhancement requests. This advises the Enhancer to include the ExecutionMetadata [1] in the returned RDF graph. Those information are also used by the default UI to show execution times of Engines. best Rupert [1] http://stanbol.apache.org/docs/trunk/components/enhancer/executionmetadata.html On Thu, Apr 25, 2013 at 9:19 PM, harish suvarna <hsuva...@gmail.com> wrote: > Sometime I observe that stanbol response time to enhance (default engine) > for even 3-4 lines is almost a minute. Is there any hibernation inside > stanbol engines or dbpedia loading, in case stanbol is not used for long > periods? > After this first request, then everything is processed fast. > > -- > Thanks > Harish -- | Rupert Westenthaler rupert.westentha...@gmail.com | Bodenlehenstraße 11 ++43-699-11108907 | A-5500 Bischofshofen
