[Paraview] pvrenderserver with MPI or without MPI?

[Berk Geveci]

> parallel. Does it still use MPI?; for example, if I had multiple
> graphics cards, say 4 cards, should I do like this?:
> % mpirun -np 8 pvdataserver
> % mpirun -np 4 pvrenderserver
> Or does it work in other way?

The render server uses mpi as well. Its main use is to support compute
server (a supercomputer for example), rendering cluster and client
scenarios. The compute server produces geometry, reduces it to N piece
(where N is the number of nodes on the render server), ships it to the
render server using N sockets and then the render server renders
geometry using distributed rendering (Ice-T over MPI).

-berk

On Mon, Jan 19, 2009 at 6:50 PM, Shigenobu Hirose <shirose at jamstec.go.jp> wrote:
> Hi Berk,
>
> Thank you for the information. It looks fine for my current purpose.
>
> But, I think I haven't yet understood how pvrenderserver works in
> parallel. Does it still use MPI?; for example, if I had multiple
> graphics cards, say 4 cards, should I do like this?:
> % mpirun -np 8 pvdataserver
> % mpirun -np 4 pvrenderserver
> Or does it work in other way?
>
> Shigenobu
>
> On Mon, Jan 19, 2009 at 23:31, Berk Geveci <berk.geveci at kitware.com> wrote:
>> Hi Shigenobu,
>>
>> I'd recommend using simple client-server and turning off remote
>> rendering from the Preferences -> Render View -> Server. This should
>> give you the same result without the complication of the render
>> server. Making use of multiple cores when there is only one graphics
>> card is a research topic for us. In the future, there will be better
>> way of doing this.
>>
>> -berk
>>
>> On Sun, Jan 18, 2009 at 10:43 PM, Shigenobu Hirose
>> <shirose at jamstec.go.jp> wrote:
>>> Hi,
>>>
>>> I've built Paraview servers on Mac OS X (10.5.6 intel x86_64) with
>>> PARAVIEW_USE_MPI=ON.
>>> I'm going to run the servers on a MacPro with 8 CPUs and with a single
>>> graphics card.
>>>
>>> I understand that
>>> % mpirun -np 8 pvserver
>>> is meaningful for data processing since it will be done in parallel
>>> using 8 CPUs. But, I think it is redundant for rendering since there
>>> is only one graphics card. In fact, when I run the above command,
>>> there appear 8 windows, but rendering seems to be done on only one of
>>> them with other 7 windows being blank.
>>>
>>> So, I guess I should launch the pvdataserver with MPI but the
>>> pvrenderserver without MPI like this:
>>> % mpirun -np 8 pvdataserver
>>> % pvrenderserver
>>>
>>> Is my understanding correct?
>>>
>>> Thank you.
>>> Shigenobu
>>> _______________________________________________
>>> ParaView mailing list
>>> ParaView at paraview.org
>>> http://www.paraview.org/mailman/listinfo/paraview
>>>
>>
>