Next: SGI logo Up: Top Previous: Introduction


Three examples have been considered for evaluation of parallel performance of 3D mesh generator -- SGI logo, chair and junction of pipes. The size of the examples corresponds to small and moderate problems and has been chosen with respect to significant limitation of memory available on Transtech Paramid machine (even small paging of memory has caused severe degradation of performance). The memory limitation has also resulted in problems with evaluation of speedup and efficiency on Paramid machine because for most of the examples there was no chance to obtained the results using only a single slave processor. In these cases, speedup was estimated in two ways. Firstly, the speedup was calculated from the idealized assumption that the speedup is ideally linear up to the minimum number of slave processors required for successful discretization of a given example. And secondly, the speedup (enclosed in parenthesis in the following) was taken from the results achieved on SP2 machine to enable approximate comparison of speedup profiles. The estimated speedup and efficiency are marked by asterisk superscript in the following. Note that only the model level of the domain decomposition have been applied in the presented examples.

The master and slaves scheme has been used as parallel computing scheme of the mesh generator. This implies that a separate processor must be allocated for master process. However, the results (speedup and efficiency) provided do not consider this master processor. In the case of IBM SP2 machine this is affordable because it has been verified that master process can be running together with one slave process on the same processor without impact on performance. However this is possible only up to 7 slave processes because of the current configuration of the SP2 machine. In the case of Transtech Paramid machine the master processor needs not to be considered as well because the master process is always running on the host machine which can be used simultaneously for other purposes (unless it will cause machine overloading).


Next: SGI logo Up: Top Previous: Introduction

Daniel Rypl