[vtkusers] Re: [vtk-developers] Deprecation questions

[Berk Geveci]

The functionality of distributing  pipeline over multiple hosts is
still there, as in data parallel pipeline. We actually have multiple
applications that rely on this functionality.
The ports are tricky because they pass pipeline passes across network.
This is the functionality I am trying to get rid of because:
* It is not commonly used 
* It is not tested well (and will not be unless someone else creates
better tests)
* It has bugs

Without ports, it will still be possible to:
* Create data parallel pipelines
* Send datasets between machines
* Use the parallel filters in VTK and ParaView

Pipeline parallelism, although an interesting research topic, is not
scalable and hard to program.
Also, a lot of the client/server functionality are now in ParaView's
ServerManager library. This library allows people to write
client/server applications and manages creating pipelines on the
server, configuring the pipeline and sending data back and forth
(mostly for local rendering on the client, probing etc). It has A LOT
of things that hide the complexity of parallel processing and
rendering,
With all of this, I think the input/output ports are pretty much
redundant. Granted, your application will no longer compile but it
might be better to switch to a better developed, tested and supported
architecture.

What do you think?

-Berk

On 8/20/05, Robert Belleman <robbel at science.uva.nl> wrote:
> Hi Berk,
> 
> I have some (old) code here that uses vtkInputPort
> and vtkOutputPort. Granted: it hasn't been used for
> a long time so I don't know if it still works, but
> I would hate to see the functionality go (i.e. the
> ability to distribute pipelines over multiple hosts).
> 
> Is there an alternative mechanism? If not, I would
> keep it in.
> 
> Best,
> -- Rob
> 
> PS: I might have missed it, but it seems you sent
>     your email only to vtk-developers. Shouldn't
>     you ask vtk-users as well?
> 
> --
> Robert Belleman, PhD, Informatics Inst.,  Faculty of Science
> Universiteit van Amsterdam, Kruislaan 403, 1098 SJ Amsterdam
> the Netherlands. Tel: +31 20 525 7510 - Fax: +31 20 525 7419
> http://www.science.uva.nl/~robbel/  -  robbel at science.uva.nl
> 
> 
> Berk Geveci wrote:
> 
> > Hi folks,
> >
> > I would like to know what you think about potentially deprecating a
> > few classes:
> >
> > In VTK/Parallel:
> >
> > * vtkThreadedController and vtkSharedMemoryCommunicator (not to be
> > confused with vtkMultiThreader): These were created as part of the
> > parallel VTK extension a few years ago. They are not well maintained
> > (hence buggy) and as far as I know unused.
> >
> > * vtkInputPort and vtkOutputPort: Again two classes that are part of
> > the parallel extensions. They allow pipelines to span networks. As
> > nice as it sounds, they are not implemented well and not maintained
> > well. They were used in writing a paper but have fallen out of use since.
> >
> > Unless there are any objections, I will remove these classes before we
> > branch for 5.0.
> >
> > In VTK/Filtering:
> >
> > Also, we are planning to deprecate vtkStructuredPoints. It will be in
> > 5.0 but will print a warning when constructed. The warning can be
> > suppressed with a compile option. vtkStructuredPoints is an empty
> > subclass of vtkImageData and has no reason to exist anymore.
> >
> > -Berk
> > _______________________________________________
> > vtk-developers mailing list
> > vtk-developers at vtk.org
> > http://www.vtk.org/mailman/listinfo/vtk-developers
> 
> 
> _______________________________________________
> vtk-developers mailing list
> vtk-developers at vtk.org
> http://www.vtk.org/mailman/listinfo/vtk-developers
>