Next: Conclusions Up: Top Previous: Parallel Computing Scheme

Example

To demonstrate the parallel performance of the algorithm an example is presented. Junction of two pipes (Fig. 2) have been discretized on IBM SP2 using up to 9 processors. It should be noted however that due to the current configuration of IBM SP2 one processor was always devoted to the master which is exploiting only small portion of the computational power of the CPU for most of the mesh generation process and consequently degrading the overall performance. Two meshes has been considered for parallel generation. The smaller one consisting of 92707 nodes was chosen so that it could have been generated on a single processor. The speedup achieved for 2 to 7 processors is depicted in Figure 3. Since the number of slaves is always one less than the number of processors actually used, the speedup is plotted separately for slaves (  - master not considered (possibly the case when master can run with one slave on the same processor)) and for all used processors (  - master considered). The larger mesh comprises 518929 nodes and the speedup achieved for 5 to 9 processors is displayed also in Figure 3. Note that the speedup for larger mesh has been calculated from the assumption that the speedup is ideally linear up to 5 processors, which is in good agreement with speedup determined from extrapolation of performance of the sequential version of this mesh generator.



Next: Conclusions Up: Top Previous: Parallel Computing Scheme

Daniel Rypl
2005-12-03