[Paraview] Reading 8GB Volume

[Jean Favre]

On 18, Jul 2008 04:45 PM, Kent Eschenberg <eschenbe at psc.edu> wrote:


>However, it seems the good old VTK XML formats using appended raw data
>are, well, "old" in that they require a 32-bit byte count which is of
>course inadequate for 8GB.

 
 
I have also abandoned the VTK XML formats because of the 32 bit byte
count; For plain, monolithic, large storage, HDF5 works very well for
me; It is of course trivial to write an Xdmf wrapper, but I have found
the performance not good at all for large parallel reads, I write my own
reader with raw HDF5 commands and I reach about 1.2 GB per sec reads; I
am happy with that; I do not use D3;
 
A further advantage is that if you write python scripts, and if you need
only hyperslabs, the pipeline is smart enough to only ask the reader to
get only the subset needed (at the difference of paraview, which always
reads the whole volume, before you have  a chance to create an Extract
Subset)
 
Jean --
Swiss National Supercomputing Centre