ctakes-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Miller, Timothy" <Timothy.Mil...@childrens.harvard.edu>
Subject Re: uima-as examples [EXTERNAL]
Date Fri, 18 Jan 2019 18:29:26 GMT
Greg - I've developed a cluster-like architecture that uses Docker-wrapped UIMA-AS components
on AWS for scalability. It's a work in progress but it might be helpful:

-----Original Message-----
From: Greg Silverman <gms@umn.edu<mailto:Greg%20Silverman%20%3cgms@umn.edu%3e>>
Reply-to: <dev@ctakes.apache.org>
To: dev@ctakes.apache.org<mailto:dev@ctakes.apache.org>
Cc: Raymond Finzel <finze006@umn.edu<mailto:Raymond%20Finzel%20%3cfinze006@umn.edu%3e>>,
Reed McEwan <rmcewan@umn.edu<mailto:Reed%20McEwan%20%3crmcewan@umn.edu%3e>>
Subject: Re: uima-as examples [EXTERNAL]
Date: Fri, 18 Jan 2019 12:23:53 -0600

Thanks Peter,
The architecture for our project (https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_nlpie_nlp-2Dadapt-2Dkube&d=DwIFaQ&c=qS4goWBT7poplM69zy_3xhKwEW14JZMSdioCoppxeFU&r=Heup-IbsIg9Q1TPOylpP9FE4GTK-OqdTDRRNQXipowRLRjx0ibQrHEo8uYx6674h&m=WEY8xYYIUiTWnZDnwU72eUiyHXNWFAi3vY9DMayfV-g&s=fvf05Pvhnq2FEnKxgYHuXibuP5Is9-bZCEE8-cbqq8M&e=,
uima-as branch under current development), relies heavily on uima-as to
work in conjunction with ActiveMQ and a home spun multiplexer/collection
processing client to do all the heavy lifting for the nlp-engines we're
using. Currently, CLAMP, and BioMedICUS both support UIMA-AS out-of-the-box
(I'm looking into MetaMap, as I type this).

To the best of my knowledge, the MQ and broker work together (at least in

Given the volume of documents we need to process and the constraint of
being tied to UIMA, UIMA-AS is the easiest option for implementing at
scale, for both speed and fault tolerance.

If anyone has done any work trying to integrate UIMA-AS into cTAKES we
would be very interested in this. Retrofitting a different solution into
our architecture at this time is not feasible.

Thanks very much!



On Thu, Jan 17, 2019 at 10:08 PM Peter Abramowitsch <pabramowitsch@gmail.com<mailto:pabramowitsch@gmail.com>>

I used a completely different approach that allows parallel but not async
processing.  Multiple [analysis engine+cas] pair objects pre-instantiated
into into a threadsafe pool running behind a web service interface. We can
fully saturate a single ctakes server process using multiple client
processes talking to that API each working synchronously and arriving at an
overall speed of 10-15 6K notes per second on a single server process.

I haven't used AS but it looks as if that middleware could have too many
moving parts for our needs.  They would generate many wakeups and context
switches adding undesired latency as a request makes its way to the
server.   I'm assuming that in AS, the broker and the MQ are separate
processes and not just in-process subsystems to the ctakes server process.
Is that right?

On Thu, Jan 17, 2019 at 4:09 PM Greg Silverman <gms@umn.edu<mailto:gms@umn.edu>>

Anyone out there developed a pipeline using UIMA-AS, as opposed to the
CPE/CPM file reader?

Thanks in advance!


Greg M. Silverman
Senior Systems Developer
NLP/IE <https://urldefense.proofpoint.com/v2/url?u=https-3A__healthinformatics.umn.edu_research_nlpie-2Dgroup&d=DwIFaQ&c=qS4goWBT7poplM69zy_3xhKwEW14JZMSdioCoppxeFU&r=Heup-IbsIg9Q1TPOylpP9FE4GTK-OqdTDRRNQXipowRLRjx0ibQrHEo8uYx6674h&m=WEY8xYYIUiTWnZDnwU72eUiyHXNWFAi3vY9DMayfV-g&s=M9wwarnfKqExquzT7isXdu4-A3-sDXx0Rar1-Y1HrlA&e=>
Cardiovascular Informatics <https://urldefense.proofpoint.com/v2/url?u=http-3A__www.med.umn.edu_cardiology_&d=DwIFaQ&c=qS4goWBT7poplM69zy_3xhKwEW14JZMSdioCoppxeFU&r=Heup-IbsIg9Q1TPOylpP9FE4GTK-OqdTDRRNQXipowRLRjx0ibQrHEo8uYx6674h&m=WEY8xYYIUiTWnZDnwU72eUiyHXNWFAi3vY9DMayfV-g&s=FdX-_52HBO3FixbIlX0HB8sdIvmjsTvd-lCM_PL3qII&e=>
University of Minnesota

 ›  evaluate-it.org  ‹

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message