Parallelizing Watergap

Watergap is the name of a program developed to simulate the distribution and flow of water on earth over several years. It is based on a global simulation model, consisting of thousands of geographical cells. Performance starts to play a critical role now, as more and more details are added to the simulation and the times required for one simulation run are increasing to several days. The program is implemented in C++.

The main goal of this thesis is to improve the performance of the program by parallelizing it. A first step could be taken by profiling the program, analyzing where it spends most of its time (possibly using several existing profilers). Afterwards, parallelization of key components should be attempted by using and comparing several parallel programming systems (such as OpenMP, MPI or hybrid solutions). Performance results can be gathered on a parallel machine.

This thesis work can possibly be complemented with paid work as a student assistant on the program. If you are interested in this topic or want more information please contact Claudia Leopold.

Jens Breitbart, 2006