You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@uima.apache.org by Steven Suppe <su...@llnl.gov> on 2007/08/24 02:12:44 UTC

CasPoolSize

Hello,

I'm curious as to the proportion for the Cas Pool Size in a CPE compared to 
the number of instances.

 From the manual...

"It could make sense to have additional CASes, though – if you had other 
multi-threaded processes that were using the CASes, other than the AEs."

What types of other processes are we talking about?  Something not 
typically used in UIMA?  Or, say, the consumers?  It seems if you have a 
'normal' pipeline of AEs and consumers, having a larger pool is 
unnecessary.  Is this true?

Thanks,
Steve

Re: CasPoolSize

Posted by Marshall Schor <ms...@schor.com>.
Steve Suppe wrote:
> Marshall,
>
> Thanks.  I hadn't thought about the Collection Reader and the CAS
> consumer.  If I have a pipeline scaled up by 3, and I have a
> multi-threaded consumer (say, 3 instances of the consumer speaking via
> Vinci to remote machines), does that mean 3 extra CAS's for the
> consumer?  So in this case 1 for the collection reader, 3 for the AEs
> and 3 for the consumers?
No - I think that the multi-deployed consumers are put into the main
pipeline and run on the pipeline thread(s).

-Marshall
>
> Thanks,
>
> Steve
>
> At 08:47 PM 8/23/2007, you wrote:
>> Steven Suppe wrote:
>> > Hello,
>> >
>> > I'm curious as to the proportion for the Cas Pool Size in a CPE
>> > compared to the number of instances.
>> >
>> > From the manual...
>> >
>> > "It could make sense to have additional CASes, though ­ if you had
>> > other multi-threaded processes that were using the CASes, other than
>> > the AEs."
>> >
>> > What types of other processes are we talking about?  Something not
>> > typically used in UIMA?  Or, say, the consumers?  It seems if you have
>> > a 'normal' pipeline of AEs and consumers, having a larger pool is
>> > unnecessary.  Is this true?
>> Yes, I think this is true.  One additional factor to be aware of - if
>> you have a pipe line scaled up by 3 (you have 3 instances of annotators
>> in the main pipe line, there will be 3 threads for this, plus an
>> additional thread for the collection reader, and one for the cas
>> consumer.  So to insure all of these are running that could be running
>> at one time, you would want to have a pool size of 5.
>>
>> -Marshall
>
>


Re: CasPoolSize

Posted by Steve Suppe <ss...@llnl.gov>.
Marshall,

Thanks.  I hadn't thought about the Collection Reader and the CAS 
consumer.  If I have a pipeline scaled up by 3, and I have a multi-threaded 
consumer (say, 3 instances of the consumer speaking via Vinci to remote 
machines), does that mean 3 extra CAS's for the consumer?  So in this case 
1 for the collection reader, 3 for the AEs and 3 for the consumers?

Thanks,

Steve

At 08:47 PM 8/23/2007, you wrote:
>Steven Suppe wrote:
> > Hello,
> >
> > I'm curious as to the proportion for the Cas Pool Size in a CPE
> > compared to the number of instances.
> >
> > From the manual...
> >
> > "It could make sense to have additional CASes, though ­ if you had
> > other multi-threaded processes that were using the CASes, other than
> > the AEs."
> >
> > What types of other processes are we talking about?  Something not
> > typically used in UIMA?  Or, say, the consumers?  It seems if you have
> > a 'normal' pipeline of AEs and consumers, having a larger pool is
> > unnecessary.  Is this true?
>Yes, I think this is true.  One additional factor to be aware of - if
>you have a pipe line scaled up by 3 (you have 3 instances of annotators
>in the main pipe line, there will be 3 threads for this, plus an
>additional thread for the collection reader, and one for the cas
>consumer.  So to insure all of these are running that could be running
>at one time, you would want to have a pool size of 5.
>
>-Marshall

Re: CasPoolSize

Posted by Marshall Schor <ms...@schor.com>.
Steven Suppe wrote:
> Hello,
>
> I'm curious as to the proportion for the Cas Pool Size in a CPE
> compared to the number of instances.
>
> From the manual...
>
> "It could make sense to have additional CASes, though – if you had
> other multi-threaded processes that were using the CASes, other than
> the AEs."
>
> What types of other processes are we talking about?  Something not
> typically used in UIMA?  Or, say, the consumers?  It seems if you have
> a 'normal' pipeline of AEs and consumers, having a larger pool is
> unnecessary.  Is this true?
Yes, I think this is true.  One additional factor to be aware of - if
you have a pipe line scaled up by 3 (you have 3 instances of annotators
in the main pipe line, there will be 3 threads for this, plus an
additional thread for the collection reader, and one for the cas
consumer.  So to insure all of these are running that could be running
at one time, you would want to have a pool size of 5. 

-Marshall