[Paraview] MPI performance and VGL error

[Moreland, Kenneth]

I will have to guess at why, say, contour is running slower.  I can think of two possibilities (but not claiming that either is necessarily correct).


 1.  Perhaps the data is not distributed.  If the reader is naive (you didn't specify what data you were reading), it could be just loading all of it on node 0.  The poor performance could simply be that the data is really being processed in serial.  Run the Process Id Scalars filter on the output of contour to make sure that the cells are distributed amongst the processes.
 2.  Some filters (contour included) need ghost cells.  When you add them, they will send a request up the pipeline to load another layer of ghost cells.  Assuming the reader supports it, the reader will usually reload the entire data set from file on each process to include the ghost cells.  There is a good chance that reading the data from disk is taking a lot longer than contouring the data.  Open the timer log (Tools -> Timer Log) and see how much time each component is taking.  If the contour filter is reported to run really fast (or missing from the timings if faster than the threshold) and the reader is running multiple times, then that is probably it.

As far as your problems with vglconnect, I am afraid I cannot be much help as I am unfamiliar with that hardware.  (Perhaps someone else on this list has run into this hardware.)  It sounds like the pvserver nodes are having trouble connecting to the appropriate X displays.  Usually this is a matter of getting your DISPLAY environment variables (or -display command line flags) correct.

-Ken


On 2/18/09 2:31 AM, "Orlando Rivera" <Orlando.Rivera at lrz.de> wrote:

Dear all,


We have compiled paraview server  with  MPICH2 and parastation ,   when
we deploy the server  , it starts with no troubles , the clients
connects  easily as well .

However , when  we do some calculations , like a contour,   it takes
much more time (~3x)    to represent the surfaces   as  when  using the
client  alone


Another thing that bothers us,  when I start the client  with  vglrun
(through vglconnect  )   We get this warning :
[VGL] ERROR: Cannot obtain a Pbuffer-enabled 24-bit FB config on the
VirtualGL
[VGL]    server.  This is most likely because the 3D X Server :1.2
[VGL]    is not configured for 24-bit color or does not have accelerated 3D
[VGL]    drivers installed.

but we have drivers installed and cmake found them .

I don't know   whether these two problems are related or not , I would like

1)  a benchmark  or a way to determinate  that the server does in fact
internal calculations when using mpi

2) How to solve the client issue with VGL

our hardware is a Sun x4600 server with 16 CPU cores (8x2 opteron )
with   SLES  10 SP2   and  four Quadro FX5500 graphics cards.


Any help is  gladly appreciated,  request  if you  need more
information form us

Thanks


--
Orlando Rivera
E-mail: Orlando.Rivera at lrz.de
web:    http://www.lrz.de
Leibniz-Rechenzentrum
Abteilung: Hochleistungssysteme
Boltzmannstrasse 1, 85748 Garching
Telefon: +49 89 35831-8827
Fax:     +49 89 35831-9700



_______________________________________________
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

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.paraview.org/pipermail/paraview/attachments/20090218/400611f5/attachment.htm>