Need Parallel processing explanations

[Spico Florin <sp...@gmail.com>]

Hello!
  We have a pipeline that uses the parallel flow controller to coordinate
parallel remote AE. They are annotating the same message.
Our pipeline runs two annotators in parallel and the outputs of them are
sent to one CAS consumer. The two annotators don't depend on each other.
  I would like to know how UIMA AS manage the parallel behavior of the two
processes when one annotator is running slower than the other?
  Will the faster annotator consumes more messages than slower one? Or they
work at the same level, and the faster one will wait until the slower one
  gets its job done, and then it consumes the next message from the
pipeline?

 I look forward for your answers.
 Thank you.

Best regards,
 Florin

Re: Need Parallel processing explanations

[Burn Lewis <bu...@gmail.com>]

You could compensate for the speed differential by scaling out the slower
annotator and using a larger pool in the aggregate so that multiple CASes
are in play together.  This doesn't change the latency of any one CAS but
would increase throughput.

~Burn

Re: Need Parallel processing explanations

[Jaroslaw Cwiklik <ui...@gmail.com>]

UIMA AS aggregate client waits for all replies to come back from Parallel
step before allowing a CAS to move on to the next step in the flow. From
the UIMA AS service point of view, services are independent and consume
messages(CASes) as soon there is capacity to do so. Faster service will
consume more messages than the slow one.

Jerry C


On Wed, Feb 1, 2012 at 10:23 AM, Spico Florin <sp...@gmail.com> wrote:

> Hello!
>  We have a pipeline that uses the parallel flow controller to coordinate
> parallel remote AE. They are annotating the same message.
> Our pipeline runs two annotators in parallel and the outputs of them are
> sent to one CAS consumer. The two annotators don't depend on each other.
>  I would like to know how UIMA AS manage the parallel behavior of the two
> processes when one annotator is running slower than the other?
>  Will the faster annotator consumes more messages than slower one? Or they
> work at the same level, and the faster one will wait until the slower one
>  gets its job done, and then it consumes the next message from the
> pipeline?
>
>  I look forward for your answers.
>  Thank you.
>
> Best regards,
>  Florin
>