MARC CLUSTER BENCHMARK Non-Linear Analysis of a Suspension Knuckle utilizing MSC Marc 2003 DDMSlide 2
HP J6700 1 CPU Solve Time: 13:26 Baseline TimeSlide 3
Linux Cluster 2 CPU's Solve Time: 5:20 Speed-Up: 2.5XSlide 4
Linux Cluster 4 CPU's Solve Time: 3:07 Speed-Up: 4.3XSlide 5
Linux Cluster 8 CPU's Solve Time: 1:51 Speed-Up: 7.3XSlide 6
Linux Cluster 16 CPU's Solve Time: 1:03 Speed-Up: 12.8XSlide 9
Benchmark Summary For the present benchmark part , doubtlessly 4 CPU's gives ideal arrangement time versus registering assets being used. Nonetheless, it ought normal that distinctive sorts of non-straight recreations will yield changing execution bends and require diverse quantities of CPU's because of the differing way of non-direct issues. While not troublesome, accommodation of Marc recreations to the bunch is not as of now as basic as "pushing go". There are various factors that must be designed to guarantee a fruitful run that yields comes about as opposed to blunders. We are taking a shot at a more proficient approach to achieve this errand. A "cookbook" approach maybe… Irregardless of the quantity of processors or the registering stage utilized, the resultant stretch esteem was dependably the same (274 Mpa) and in a similar area. No deviation in exactness was recognized. There appears to at present be a "bug" in the Linux form of the Mentat Pre-handling programming. The HP rendition must be utilized to make the information decks for the group runs. MSC has been cautioned to this and is taking a shot at revising the issue. The Linux Networx equipment has so far performed faultlessly in the majority of the classifications that it has been tried against. They have additionally been exceptionally proficient in loaning backing to MSC to determine programming issues in the Marc application.