Improve the performance of multiplier as high as possible in the version 2.2.2 of UIMAAS

[feixiang <fe...@gmail.com>]

  I test my program and find that the multiplier is the bottleneck.I find 
the multiplier's idle time is always 0.So I want to improve the performance
 by speeding up the multiplier. The tutorial says,Only one remote can be a 
remote CAS Multiplier,In version 2.2.2, and that remote can only have one 
instance.I still deploy the multiplier 5 instances in 1 node.The performance 
improves, 2 times faster than 1 instance,not 5 times.What does the result tell
to me? 
  I dont't konw the relationship between CASpool and casMultiplier poolSize. 
The tutorial says,the actual CAS pool size is the value in the deployment 
descriptor, plus the value in custom CM code, minus 1.That assumes there is 1 
instance.What about 5 instances?
  So in the version 2.2.2,how can I improve the performance of multiplier as
 high as possible?Eddie Epstein said,"getEmptyCas() will block until a free CAS
 is available in the CasPool."
  So how can I avoid the "getEmptyCas()" is blocked?If I deloly 5 instances of 
multiplier,How do I assign the size of CASpool and casMultiplier poolSize?
  I　am very grateful if you help me.

Re: Improve the performance of multiplier as high as possible in the version 2.2.2 of UIMAAS

[Eddie Epstein <ea...@gmail.com>]

First of all, version 2.2.2 does not reliably support scaled out remote
CasMultipliers, but the current version 2.3.0 does. Please move to
the new version.

>  I dont't konw the relationship between CASpool and casMultiplier poolSize.
> The tutorial says,the actual CAS pool size is the value in the deployment
> descriptor, plus the value in custom CM code, minus 1.That assumes there is 1
> instance.What about 5 instances?
I don't believe any adjustment is made for the number of CM instances.
As of now the user will have to calculate the right size to set the CasPool.

Also note that an aggregate with a remote CasMultiplier delegate requires
configuration of a "ShadowCasPool". The shadow CasPool size is the number
that will be distributed by the aggregate.

If you would share the block diagram of your intended configuration, either
on this list or privately, we would be in a better position to help.

Eddie

On Tue, May 25, 2010 at 11:25 PM, feixiang <fe...@gmail.com> wrote:
>
>  I test my program and find that the multiplier is the bottleneck.I find
> the multiplier's idle time is always 0.So I want to improve the performance
>  by speeding up the multiplier. The tutorial says,Only one remote can be a
> remote CAS Multiplier,In version 2.2.2, and that remote can only have one
> instance.I still deploy the multiplier 5 instances in 1 node.The performance
> improves, 2 times faster than 1 instance,not 5 times.What does the result tell
> to me?
>  I dont't konw the relationship between CASpool and casMultiplier poolSize.
> The tutorial says,the actual CAS pool size is the value in the deployment
> descriptor, plus the value in custom CM code, minus 1.That assumes there is 1
> instance.What about 5 instances?
>  So in the version 2.2.2,how can I improve the performance of multiplier as
>  high as possible?Eddie Epstein said,"getEmptyCas() will block until a free CAS
>  is available in the CasPool."
>  So how can I avoid the "getEmptyCas()" is blocked?If I deloly 5 instances of
> multiplier,How do I assign the size of CASpool and casMultiplier poolSize?
>  I　am very grateful if you help me.
>
>
>