SC²S Colloquium - June 4, 2008
|Time:||3 pm, s.t.|
15:00 Uhr - Yevgeniy Terekhin: Modern compute clusters: from hardware to application
The task of my work is parallelization and optimization of an algorithm employed in MD simulations. For the parallelization, message passing approach. In the implementation of the parallel code, the MPI is used. The focus of the parallelization is mainly to build a Cartesian topology, the exchange of data within it, and others. To optimize the performance of the code methods such as loop unrolling, variable renaming etc. are used. To optimize communication within the processes I use asynchronous transfer of data. The current status of the implementation and results are presented.