[Users] VisIt HDF5 AMR Vectors and Streamline Rendering

Gregory Colten colten1 at illinois.edu
Tue Feb 12 11:51:16 CST 2013

Here is an overview of our goal:

We are trying to visualize vectors in the default build of VisIt 2.6.0. Our
data is in 3 sets of HDF5 files, one for each of the x, y, and z
components. And for each component, there are 48 files (chunked as the
output of each of the 48 processors). The simulation uses Cactus AMR grids.
The total size of all of these files is roughly 30 GB, and each one
individually is roughly 210 MB. We find it either takes an absurd amount of
time or doesn't work at all.

We have found one method of visualizing these vectors, but it is
outrageously slow.

Working method:
This method uses 100% of one CPU and ~57% of the RAM (of 8 GB in total). It
takes nearly 25 minutes to render one timestep when displaying just the
Vectors module.

Here's what the method includes:

Database correlation between the three components (Correlation Method: Time)

Data-Level comparisons:
Between meshes in two different databases
*Position*-Based CMFE
Simply place it on the target mesh

We modified the expressions to place angled brackets around <Carpet AMR

Made an expression that puts these three components into a vector
ex {<MHD_EVOLVE--mhd_st_x>,By_position,Bz_position}

Almost-working method:
The only change made to the above process is in the Data-Level Comparisons:
(Connectivity based instead of position based)
Between meshes in two different databases
*Connectivity*-Based CMFE
Simply place it on the target mesh

This runs significantly more quickly on the first timestep in the data. But
when we try to advance to the next time step in the database correlations,
we get this error:
ERROR: Vector: ()

viewer: The databases cannot be compared because they have a different
number of domains.

We believe this to be a result of the moving AMR boxes, but we cannot seem
to work around this. The components of the vector should each have the
exact same positions as each other at all times, but VisIt seems to
disagree with this.

Our thoughts to resolve this primarily center around finding some manner to
make the Connectivity based CMFE function. If that's not possible, we can
try and see if building a parallel version of Visit 2.6.0 might help speed
this up. Any advise on where to go from here?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.einsteintoolkit.org/pipermail/users/attachments/20130212/1e1f979a/attachment.html 

More information about the Users mailing list