| Popis: |
Over the last years, a lot of effort has been spent on the development and implementation of parallel computing techniques for Modelica models. Especially the field of automatic parallel computation of the state derivatives has been studied intensively.First results of the project HPC-OM have shown, that a significant speedup for simulation models of the fluid domain can be achieved by parallel computation of the BLT structure blocks. But it was noticed, that the theoretical speedups are much higher than the measured ones.In this work, a performance analysis approach for the C++ based OpenModelica simulation runtime is presented. It is shown that the programmer can get a detailed insight into the behaviour of simulation runs with the help of the performance analysis tool Score-P. The gained knowledge from the visualization of workflow issues and performance bottlenecks can be used to reduce the computation time significantly. More precisely, it is shown that the influence of false sharing and I/O operations can be improved with an optimized implementation in order to improve the performance. |
