The default mechanism Vinci uses to send requests to registered servers
is "random".  This has the property that not necessarily all service
instances get used.  There is an alternative "round-robin" mechanism
that can be specified, which will use all the servers.  It is specified
in the CPE descriptor as a <deploymentParameters> element called
"services-access".

See section 3.6.1.5: 
http://incubator.apache.org/uima/downloads/releaseDocs/2.2.0-incubating/docs/html/references/references.html#ugr.ref.xml.cpe_descriptor.descriptor.cas_processors.individual

Some of us are working on a more general approach to remoting and
parallelization, described on our wiki, here:
http://cwiki.apache.org/UIMA/uimaasdoc.html

-Marshall

candel wrote:
> Hello,
>
> I am studying remote services with UIMA. Especially with Vinci services.
> I did several tests to understand how it was working.
> I was testing several parameters :
>    numInstances (vinci service deployment descriptor)
>    casPoolSize and processingUnitThreadCount (collection reader, CPE
> descriptor)
> My goal was to perform parallelism by opening several identical's
> services.
> But i remarked that not all the services were running even if i
> changed these parameters.
> (even after deleting VNS.counters and VNS.services files)
> My questions are:
>    1/ Is it a good way to do parallelism with remote services?
>    2/ How UIMA deals with parallelism?
> The best of all for me would be further documentation on this subject
> (I got Appache-UIMA docs)
> Thanks a lot,
>
> Sophie
>
>
>

Reply via email to