# Parallel Computations

The parallel solution within Elmer utilizes the Message Passing Interface (MPI) standard for inter-process communication. The distribution of load to the different processes is based on domain decomposition. This implies that the mesh is split into N parts (if the parallel computation consists of N processes).

- An introduction example: parallelexample.tar.gz

## Using the SaveLine solver for parallel computations

The SaveLine solver module allows the extraction of one or multiple data sets along lines through the complete geometric domain. Such a line may cross concurrently calculated partition boundaries. The data output from the SaveLine solver is dumped into different data sets, named accordingly *anyhow.dat.0*, *anyhow.dat.1*. etc., the end number representing the calculational domain of the parallel run.

The user or rather the evaluation program then has to take care that the data of the individual calculation domains have to be recombined into one display. An example is given for a quad-core calculation of the time dependent temperature profiles in a brick wall:

The image was obtained from GnuPlot, with the three different colors representing the data output from three of the four calculational domains. The fourth did not contribute data to the extracted line.