Parallel Computation of the Topological Skeleton of Vector Fields

Vector fields occur in many of the problems in science and engineering. In combustion processes, for instance, vector fields describe the flow of the gas. This process can be enhanced using vector field visualization techniques. Also, wind tunnel experiments can be analyzed. An example is the design of an air wing. The wing can be optimized to create a smoother flow around it. To analyze such kind of datasets topological methods that clearly show the whole structure of the vector field in one picture are a very good tool. During the last years, many extensions were proposed for this method. In addition to standard topological methods we also detect closed streamlines since they are a topological feature that completes the topological analysis. To accelerate the computation of such a topological analysis we developed a parallel method to reduce computational time. Therefore, we spread the computation of the separatrices of the topological skeleton to clients of a computer cluster. To test our implementation we use a numerical simulation of a swirling jet with an inflow into a steady medium. We built two different Linux clusters as parallel test systems where we check the performance increase when adding more processors to the cluster. We show that we have a very low parallel overhead due to the neglectable communication expense of our implementation