[Paraview] Parallel ParaView: inexplicable results.

Mon Sep 28 12:27:24 EDT 2009

Ok, I see what you mean.

This is actually a very good guess. I _did_ try to run the ProcessId filter before the D3 filter, but I noticed that most (roughly 95%) of all cells were allocated to process 0, and the rest scattered over the remaining nodes with less and less for increasing process IDs. Therefore, I swapped the order of the filters thinking this would correct matters. But it only did so in case 1.

-Martin

--- Den man 28/9/09 skrev Moreland, Kenneth <kmorel at sandia.gov>:

> Fra: Moreland, Kenneth <kmorel at sandia.gov>
> Emne: Re: [Paraview] Parallel ParaView: inexplicable results.
> Til: "Martin Aunskjær" <maaudk at yahoo.dk>
> Cc: "paraview" <paraview at paraview.org>
> Dato: mandag 28. september 2009 16.52
> 
> 
> Re: [Paraview] Parallel ParaView: inexplicable
> results.
> 
>  
> (Replying back to ParaView list).
> 
> 
> 
> I was not suggesting that your file system has anything to
> do with D3.  D3 never accesses your file system.
>  What I was saying was that the only thing I can think
> of that might be wrong is that perhaps there is something
> wrong with the reader.  Perhaps the reader is reading
> the same data on every processes instead of reading a piece
> of the data on each process.  Then, perhaps, D3 is
> having trouble realizing that all the data is replicated.
>  This is just speculation, though.
> 
> 
> 
> What does it look like when you run the Process Id filter
> before D3?
> 
> 
> 
> -Ken
> 
> 
> 
> 
> 
> On 9/25/09 6:28 AM, "Martin Aunskjær"
> <maaudk at yahoo.dk> wrote:
> 
> 
> 
> Thank you
> for your reply.
> 
> 
> 
> We did build PV 3.6.1 with OSMesa using Mesa-7.5.1 but were
> unable to make it work. As far as I remember PV would not
> launch on remote nodes, and we gave up on figuring out why.
> I do not recall the specifics of the problem. Also, since
> every node in the cluster does have rendering hardware,
> OSMesa should be a last resort anyway.
> 
> 
> 
> The data is initially replicated everywhere in that it is
> present in a folder on the node where the client runs and
> this folder is exported via NFS to all other nodes. I do not
> understand why this should confuse D3, at least I would
> expect any confusion to be the same in either case ?
> 
> 
> 
> Martin Ausnkjaer
> 
> 
> 
> --- Den tors 24/9/09 skrev Moreland, Kenneth <kmorel at sandia.gov>:
> 
> 
> 
> > Fra: Moreland, Kenneth <kmorel at sandia.gov>
> 
> > Emne: Re: [Paraview] Parallel ParaView: inexplicable
> results.
> 
> > Til: "Martin Aunskjær" <maaudk at yahoo.dk>, "paraview at paraview.org" <paraview at paraview.org>
> 
> > Dato: torsdag 24. september 2009 18.02
> 
> >
> 
> >
> 
> > Re: [Paraview] Parallel ParaView: inexplicable
> 
> > results.
> 
> >
> 
> > 
> 
> > Why is OSMesa not an option?
> 
> >  Mesa 3D is free and it compiles just about
> 
> > everywhere.
> 
> >
> 
> >
> 
> >
> 
> > The IceT errors you are getting are actually probably
> 
> > caused by you having different resolutions on the
> local X
> 
> > servers.  Specifically, the X server on node 0
> probably
> 
> > has a larger resolution than at least one of the other
> X
> 
> > servers.
> 
> >
> 
> >
> 
> >
> 
> > I cannot think of a reason why D3 would behave
> differently
> 
> > between case 1 and 2.  As far as the filter
> execution
> 
> > is concerned, the two cases should be the same.
>  Could
> 
> > it be that your data is initially replicated
> everywhere and
> 
> > that D3 is somehow not able to combine co-located
> points?
> 
> >
> 
> >
> 
> >
> 
> > -Ken
> 
> >
> 
> >
> 
> >
> 
> >
> 
> >
> 
> > On 9/23/09 11:36 PM, "Martin
> Aunskjær"
> 
> > <maaudk at yahoo.dk> wrote:
> 
> >
> 
> >
> 
> >
> 
> > I am getting
> 
> > the infamous warning
> 
> >
> 
> >
> 
> >
> 
> > "Display is not accessible on the server side.
> 
> >
> 
> > Remote rendering will be disabled."
> 
> >
> 
> >
> 
> >
> 
> > whenever I start PV in client/server mode. This is
> because
> 
> > the remote pvservers cannot connect to the local X
> servers.
> 
> > The only way I have been able to get rid of it is to
> 
> > physically log in to each remote node before I
> establish the
> 
> > client/server connection (OSMesa not an option). Now,
> I can
> 
> > either choose to do that (case 1), or I can choose to
> accept
> 
> > the above warning and tolerate that only data
> processing
> 
> > will be parallel - rendering will be done locally on
> the
> 
> > client (case 2).
> 
> >
> 
> >
> 
> >
> 
> > I have observed that the following procedure produces
> 
> > different results in case 1 and 2:
> 
> >
> 
> >
> 
> >
> 
> > Start client/server using reverse connect.
> 
> >
> 
> > Load data set.
> 
> >
> 
> > Apply D3 filter.
> 
> >
> 
> > Apply ProcessId filter.
> 
> >
> 
> > Adjust color map resolution to equal number of
> pvserver
> 
> > nodes.
> 
> >
> 
> > Color data set with process id.
> 
> >
> 
> >
> 
> >
> 
> > Case 1:
> 
> >
> 
> > -------
> 
> >
> 
> > The whole data set is colored with process 0 color.
> 
> >
> 
> > The pvserver continuosly spits out:
> 
> >
> 
> > ICET,0:ERROR: Sizes of src and dest do not agree.
> 
> >
> 
> >
> 
> >
> 
> > Case 2:
> 
> >
> 
> > -------
> 
> >
> 
> > Data set colored as expected showing equal
> distribution of
> 
> > data between nodes.
> 
> >
> 
> > pvserver spits out:
> 
> >
> 
> > No protocol specified.
> 
> >
> 
> >
> 
> >
> 
> > In both cases Remote Rendering Threshold is on and set
> to 0
> 
> > MB.
> 
> >
> 
> >
> 
> >
> 
> > The ICET error is likely caused by different screen
> 
> > resolutions on servers and client. It is not possible
> for me
> 
> > to have the same resolution everywhere.
> 
> >
> 
> >
> 
> >
> 
> > The 'No protocol specified' is due to the
> missing X
> 
> > server connection, so this is expected.
> 
> >
> 
> >
> 
> >
> 
> > Why is case 1 different from case 2 - is it because of
> the
> 
> > ICET error ?
> 
> >
> 
> >
> 
> >
> 
> >
> 
> >
> 
> > A few details:
> 
> >
> 
> > --------------
> 
> >
> 
> > ParaView 3.6.1 Release build from sources.
> 
> >
> 
> > Data set size 111 MB.
> 
> >
> 
> > VTK_MPI_MAX_NUMPROCS:STRING=32
> 
> >
> 
> >
> 
> >
> 
> >
> 
> >
> 
> >
> 
> >
> 
> >
>       ___________________________________________________________
> 
> >
> 
> > Skal du købe ny bil? Sammenlign priser
> på
> 
> > brugte biler med Kelkoo og find et godt tilbud! - Se
> mere
> 
> > her http://dk.yahoo.com/r/pat/mmb
> 
> >
> 
> > _______________________________________________
> 
> >
> 
> > Powered by www.kitware.com
> 
> >
> 
> >
> 
> >
> 
> > Visit other Kitware open-source projects at http://www.kitware.com/opensource/opensource.html
> 
> >
> 
> >
> 
> >
> 
> > Please keep messages on-topic and check the ParaView
> Wiki
> 
> > at: http://paraview.org/Wiki/ParaView
> 
> >
> 
> >
> 
> >
> 
> > Follow this link to subscribe/unsubscribe:
> 
> >
> 
> > http://www.paraview.org/mailman/listinfo/paraview
> 
> >
> 
> >
> 
> >
> 
> >
> 
> >
> 
> >
> 
> >
> 
> >
> 
> >
> 
> >    ****
> 
> >      Kenneth Moreland
> 
> >
> 
> >     ***
> 
> >      Sandia National Laboratories
> 
> >
> 
> > ***********  
> 
> >
> 
> > *** *** ***  email: kmorel at sandia.gov
> 
> >
> 
> > **  ***  **  phone: (505) 844-8919
> 
> >
> 
> >     ***
> 
> >      web:   http://www.cs.unm.edu/~kmorel
> 
> >
> 
> >
> 
> >
> 
> >
> 
> > 
> 
> >
> 
> >
> 
> >
> 
> 
> 
> 
> 
>       ________________________________________________________
> 
> Audi, Fiat, Peugeot, Skoda, Porsche, Toyota, Ford - Kelkoo
> har brugte biler til en hver smag! Klik her for at
> sammenligne priser.(http://dk.yahoo.com/r/pat/mmb)
> 
> 
> 
> 
> 
> 
> 
> 
> 
>    ****
>      Kenneth Moreland
> 
>     ***
>      Sandia National Laboratories
> 
> ***********  
> 
> *** *** ***  email: kmorel at sandia.gov
> 
> **  ***  **  phone: (505) 844-8919
> 
>     ***
>      web:   http://www.cs.unm.edu/~kmorel
> 
> 
> 
> 
>  
> 
> 
> 

      Find din nye laptop på kelkoo.dk. Se de gode tilbud her - http://dk.yahoo.com/r/pat/mm