SC09Cluster
SC09 Cluster challenge (VisIt portion):
You will need to do the following:
- Build VisIt
- Generate a toy data set
- Run a script that makes a movie
Hank Childs will be at SC09 and will offer assistance if you run into trouble (hankchilds AT gmail DOT com).
Build VisIt
- You almost certainly want to use the build_visit script.
- If you run the GUI version, you will not need any of the optional third party database libraries, so don't select those.
- Alternatively, you can run: "./build_visit --console" and that should just work.
- This should build the necessary third party libraries and also VisIt itself (in serial).
It is possible to tell build_visit how to build in parallel, but that sometimes goes wrong. To get the parallel version:
- cd into the "/path/to/visit/src" directory.
- run
env CXXFLAGS="-I/path/to/mpi/include -O2" MPI_LIBS="-L/path/to/mpi/lib -lmpi" ./configure --enable-parallel # Your MPI libs and paths will likely vary make
- Also see ParallelPorting for more information about building and running VisIt in parallel.
Making a toy data set
We will make a large data set from our toy "noise" data set. We will use the program "visitconvert", which came with VisIt. You can the data set here
- /path/to/visit/src/bin
- visitconvert noise.vtk noise BOV -target_chunks 512 -target_zones 8000000000
- This will create a data set named "noise.bov" with 512 chunks and eight billion zones. If you have more than 512 processors, you should increase "target_chunks".
- It will take a while to run. You should see noise_%04d.bof.gz files being put in the file system while it runs.
- Note: I have no idea how big the clusters will be. One billion cells for every 16 cores is probably a good rule of thumb. (So reduce the 8 billion if you have less than 128 cores.)
Script
Hank will post a script by Monday that exercises VisIt on this large data set. This script will generate a movie. Send Hank a note (hankchilds AT gmail DOT com) if you want to be informed the moment the script gets posted.
