[owl-s] Thread safety of loading ontologies and performing queries?

[Jim Andreou]

2009/6/16 Thorsten Möller <Thorsten.Moeller at unibas.ch>

>
> > It would be nice to be able to download URIs from the network
> > concurrently
> This can be achieved indirectly. Start with one KB per thread and load
> the ontologies per thread in its "own" KB. After all threads finished
> then you can merge KBs into one (large) one using
> OWLKnowledgeBase.load(...), of course, using just one thread. Note that
> you might want to skip the base ontology of every KB, cf.
> OWLKnowledgeBase.getBaseOntology(). The load operation should be fast
> since it basically adds references of models rather than copying
> statements. Give it a try.
>
> Cheers,
> Thorsten
>
> [1] http://jena.sourceforge.net/how-to/concurrency.html
> [2] http://clarkparsia.com/pellet/faq/jena-concurrency/
>

Done that, works out nicely (thanks :)). For 6 test (non-local) URIs,
serially took 22sec, parallely less than half. (I know the numbers are
meaningless as-is, but I merely say it's while-worthy).
If we had:

- Some simple utilities that turn a URI/URL/InputStream(/String, why not) to
an OWLOntology
- A callback that takes OWLOntology (but this really belongs in Java :(
instead of reinventing yet another Handler<T> callback interface)

Then this could be nicely abstracted out into a reusable utility.

Here is a sketch of my code:

    static void loadServices(Handler<OWLOntology> handler, Collection<URI>
services, ExecutorService executor) throws IOException, InterruptedException
{
        CompletionService<OWLOntology> completionService = new
ExecutorCompletionService<OWLOntology>(executor);

        for (final URI service : services) {
            completionService.submit(new Callable<OWLOntology>() {
                public OWLOntology call() throws Exception {
                    return OwlsUtils.parseURI(service); //are there such
methods available? It creates a dummy OWLKnowledgeBase for each call.
                }
            });
        }

        try {
            for (int i = 0; i < services.size(); i++) { //assuming
"services" is not modified under the hood
                OWLOntology ontology = completionService.take().get();
                handler.handle(ontology); //this loads the ontology in my kb
with a write lock
            }
        } catch (ExecutionException e) {
            Throwable cause = e.getCause();
            if (cause instanceof IOException) {
                throw (IOException)cause;
            }
            throw new RuntimeException(cause); //a bit sloppy
        }
    }

//this overload probably is not a good idea
    static void loadServices(Handler<OWLOntology> handler, Collection<URI>
services) throws IOException, InterruptedException {
        ExecutorService executor = Executors.newCachedThreadPool();
        loadServices(handler, services, executor);
        executor.shutdown();
    }
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.mindswap.org/pipermail/owl-s/attachments/20090616/3b05b98e/attachment.html