This should be fixed in CVS now (at least for image data). The requested piece was not passed down into request data this every node was trying to produce everything. Also, the output image data was not using dimension and extent to specify the global and local regions properly.
cheers, Dave On Tue, Aug 26, 2008 at 10:23 AM, Jean Favre <[EMAIL PROTECTED]> wrote: > Eric E. Monson wrote: >> >> We use the Xdmf wrappers for our HDF5 for now. > > I am still encounteering major difficulties with Xdmf (cvs version of > yesterday). Speed and Memory. When reading in parallel for example, the Xdmf > reader attemps to allocate memory for the WHOLE grid on every node > (independent of the number of pvserver nodes) causing memory crashes. There > must be a problem reading the Extents. Using straight HDF5, I pass the > Extents as hyperslabs and it works fine. I can read 2048^3 in parallel. For > smaller grids, if memory is not a problem, the Xdmf wrapper works too, still > considerably slower. > > Jean -- Swiss National Supercomputing Centre > > _______________________________________________ > ParaView mailing list > [email protected] > http://www.paraview.org/mailman/listinfo/paraview > _______________________________________________ ParaView mailing list [email protected] http://www.paraview.org/mailman/listinfo/paraview
