[Paraview] Parallel rendering of XDMF unstructured data

[sj R]

Hello,

 Hello,

 I have an unstructured mesh, with 2 million points, in a single
'HDF5/XDMF' file format that I am trying to visualize on our clusters. The
goal is to perform remote parallel rendering on our cluster ( 16 processors
per node and 64 GB memory) nodes using a client-server model.

Currently I load this file on a single node with 16 processes, apply the D3
filter to it, followed by the application of a 'Clip' filter.  This dataset
contains time-steps and when I try to animate it over time, the memory
usage on the remote compute node increases with each time-step and this
eventually causes 'pvserver' to crash.

However, I can partition the data using D3 and write it out as '*.pvtu'
files. When I restart ParaView and read the '*.pvtu' files back in and
perform the same rendering tests, the performance seems to scale pretty
well and memory usage is kept to a minimum.

Is this something to do with the XDMF reader ? Is the D3 filter not being
applied correctly ?

Thanks,
Srijith Rajamohan
Computational Scientist
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/paraview/attachments/20150120/d29d2139/attachment.html>