| To develop and implement new algorithms in order to improve the parallel scaling of the GROMACS molecular dynamics code. The current algorithm implemented in GROMACS is extremely fast on single processors or small symmetric multi processing (SMP) machines, but does not scale very well on clusters and larger SMP machines. This is a problem as the run time for a simulation system of a given size can not be reduced very much by adding more computer nodes. We are applying for funds to employ a PostDoc in the field of computer science or computational physics to implement new algorithms that will improve the scaling. We propose the use of shared-memory multi-threading within modern multi-core nodes, a medium number of such nodes can parallelize very efficiently (> 90%) using message-passing calls. In addition, we will develop distributed computing techniques like multi-scaling to enable parallelization of small systems to thousands of nodes. |
