Besides the code from worksheet 1, the following additional helper files are given:
There's a small MPI demo available here. It shows the strucuture of any MPI application.
|June, 20th||Parallelization and Implementation|
According to the worksheet, each node writes vtk files of its own, i.e. the sequential visualisation environment is used for the parallel code, too. The results from the different nodes then have to be merged within the Paraview environment. There are different options to obtain merged results and nice pictures.
First, load all result files into Paraview. Translate the results along the z-axis. To improve the readability of the results you might switch to a wireframe representation or reduce the opacity. The figure below gives an example for an artificial problem.
Each node writes only the data of the domain partitions he is responsible for. If there is a default value for all the other parts of the domain, Paraview also can mask the data and visualise only the part of the data that does not equal the default value. Use Paraview's Filters/Common/Threshold option, e.g..
As soon as you are sure that your results are correct, perhaps the nicest way to end up with a nice visualisation is to modify the visualisation operations themselves. There is no need to write all the data all the time. Instead, make the visualisation operations accept arguments about their subdomain and plot only the subgrid that belong the the node's parition.
email@example.com, Mehl, 24.06.2008