[Paraview] Tiling configuration?

[Phillip Ezolt]

Ok,

>This is the first mpi implementation that does not use the same order
> the user specifies.

It is very possible that I am using it wrong (or found a bug).  I'm
going to ask around inside of HP to figure out what's going on.

Thanks for the help,
--Phil

On Tue, 2005-05-24 at 15:24 -0400, Berk Geveci wrote:
> This is the first mpi implementation that does not use the same order
> the user specifies. As far as I know, this can't be done in the
> configuration file, sorry. I don't even know if Ice-T lets one do
> this. If it does, adding something to the configuration file wouldn't
> be too hard.
> 
> On 5/24/05, Phillip Ezolt (exchange) <Phillip.Ezolt at hp.com> wrote:
> > Amy & Berk,
> > 
> > > The first machine in the list you pass to MPI is used as the upper
> > > left  tile.  Subsequent machines fill in a single row, and then the
> > > rows are
> > > filled from top to bottom.  For your 2 x 2 tiled display to give the
> > > appropriate results, the machines should be controlling the tiles as shown
> > > below:
> > 
> > This mapping can not be changed within a configuration file?
> > 
> > Something like (which would reverse the order of the tiles):
> > 
> > mpi_node_id     tile_id
> > -----------------------
> > 0                  3
> > 1                  2
> > 2                  1
> > 3                  0
> > 
> > ...
> > 
> > It looks like if I can whip my MPI job launch into shape, I can work
> > around this. (I am by no means an MPI expect, and I am looking into how
> > to solve this.)  I haven't yet figured out how to use mpirun to specify
> > the exact order of nodes...  When I reverse list,
> > (tigger6,tigger5,tigger4,tigger3) the MPI ranks are the same as when I
> > launch it with (tigger3,tigger4,tigger5,tigger6).
> > 
> > In any event, I was hoping that we could use paraview to set the tile
> > layout.  However, it looks like I have to make mpirun put the correct
> > MPI ranks on the proper display tiles.   I'll attack that.
> > 
> > Thanks for the help,
> > --Phil
> > 
> > On Tue, 2005-05-24 at 14:09 -0400, Amy Squillacote wrote:
> > > Hi Phil,
> > >
> > > Which machines are displaying their results on which tile?
> > >
> > > The first machine in the list you pass to MPI is used as the upper left
> > > tile.  Subsequent machines fill in a single row, and then the rows are
> > > filled from top to bottom.  For your 2 x 2 tiled display to give the
> > > appropriate results, the machines should be controlling the tiles as shown
> > > below:
> > >
> > > tigger3 tigger4
> > > tigger5 tigger6
> > >
> > > This issue is the same regardless of whether you use a .pvx file.
> > >
> > > - Amy
> > >
> > > At 01:49 PM 5/24/2005, Phillip Ezolt wrote:
> > > >Hi,
> > > >         I have a 2x2 tile display that I want to tile paraview (v2.0.1)
> > > >across.
> > > >
> > > >I am running in client/server mode, with 4 servers running in an MPI job
> > > >within a cluster, and a single client running outside the cluster.
> > > >
> > > >Servers are launched with the following:
> > > >
> > > >/opt/hpmpi/bin/mpirun -e DISPLAY=:0 -srun -w
> > > >"tigger3,tigger4,tigger5,tigger6" pvserver -rc --client-host=catan -
> > > >tdx=2 -tdy=2
> > > >
> > > >The client is launched with the following:
> > > >env PV_SEPERATE_RENDER_WINDOW=1 ./pvclient --reverse-connection -tdx=2 -
> > > >tdy=2
> > > >
> > > >Visualization works fine, but, unfortunately, the tile order is messed
> > > >up. The visualization on each of the tiles looks right, but the
> > > >tiles are in the completely wrong order. (It kinda looks like a messed-
> > > >up sliding-block puzzle. )
> > > >
> > > >How do I tell paraview the layout of tiles?
> > > >
> > > >I've seen some reference to a pvx file, but it is not documented (that I
> > > >could find) in the book.
> > > >
> > > >Any help is greatly appreciated.
> > > >
> > > >Thanks,
> > > >--Phil
> > > >--
> > > >Phillip Ezolt <Phillip.Ezolt at hp.com>
> > > >
> > > >_______________________________________________
> > > >ParaView mailing list
> > > >ParaView at paraview.org
> > > >http://www.paraview.org/mailman/listinfo/paraview
> > >
> > > _______________________________________________
> > > ParaView mailing list
> > > ParaView at paraview.org
> > > http://www.paraview.org/mailman/listinfo/paraview
> > _______________________________________________
> > ParaView mailing list
> > ParaView at paraview.org
> > http://www.paraview.org/mailman/listinfo/paraview
> >
-- 
Phillip Ezolt <Phillip.Ezolt at hp.com>