Mailing-List: contact user-help@uima.apache.org; run by ezmlm
Precedence: bulk
Reply-To: user@uima.apache.org
Date: Wed, 20 May 2015 15:10:44 +0200
From: Petr Baudis <pasky@ucw.cz>
To: user@uima.apache.org
Subject: Re: Multi-threaded UIMA ParallelStep
Message-ID: <20150520131043.GK2760@machine.or.cz>
References: <20150520012720.GC2760@machine.or.cz>
 <CAGRYgEOd=w+BQF6hpWd6-nP6ej8-0SECq0AzVGXYcGSJc3g6EQ@mail.gmail.com>
 <20150520124950.GJ2760@machine.or.cz>
 <CAGRYgEPAm8kSSeuJJ-ax023=V069RQr8S2uuvzJw9N78AJO2vQ@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: 
 <CAGRYgEPAm8kSSeuJJ-ax023=V069RQr8S2uuvzJw9N78AJO2vQ@mail.gmail.com>
User-Agent: Mutt/1.5.23 (2014-03-12)

  Hi!

On Wed, May 20, 2015 at 09:01:10AM -0400, Eddie Epstein wrote:
> Right about the flow controller. That's where UIMA-AS comes in. Assuming
> that the CM has a casPool with enough CASes, and the aggregate is deployed
> asynchronously, then each delegate will be running in its own thread and
> can be processing CASes in parallel.
> 
> The ASB is a single-theaded controller used for deployment of synchronous
> aggregates.

  Hmm, I think I begin to see how it should work.  Thanks for the
explanation!

  However, (before actually trying) it still seems to me to be much
easier to rewrite a piece of the stock ASB than use UIMA-AS with complex
pipeline construed by UIMAfit...  So I think I will try that first (and
report back).

> Is the intention here to use parallel processing to reduce latency for a
> interactive application or to increase throughput for batch processing? For
> throughput, why not just deploy the entire pipeline single-threaded and
> then run multiple pipeline instances in separate threads? UIMA-AS would do
> this by specifying N instances of a synchronous top-level aggregate.

  The intention is to do both, latency is also important for me
as I have an interactive user interface for the pipeline.

				Petr Baudis