The proposed work aims to enhance performance of molecular dynamics (MD) simulation code using various high-performance computing (HPC) approaches. The two-dimensional (2D) legacy code is parallelized using message-passing interface (MPI). Parallelization strategies when deployed with HPC platform, the performance and scalability improve with reduction in required computational time. Simulation experiments included two different numbers of atoms deployment keeping step size, time step, initial and boundary condition constant. Various profiling tools have been applied for identifying the hot spots that consume most of the execution time in the code. MD code is optimized employing following four approaches namely (1) force decomposition, (2) force decomposition with data organization, (3) intra- and inter-force decomposition with data organization and (4) intra- and inter-force decomposition with data organization and grid management. The output of these approaches is tested for the required accuracy by comparing its results with original standard MPI parallelized code. Simulation results for these approaches are found satisfactory from performance aspect. A comparative study is carried based on various performance metrics like execution time, speedup ratio and efficiency with multiprocessors. These approaches, when deployed on various platforms, are found better than standard MPI parallelized code except for the data organization approach. When the code is reformed implementing all approaches, the maximum speedup is found in the range of 2.5–4.5 times based on use of number of processors. Enhancement of code performance by saving computation time helps to solve the large-scale problems more efficiently.
CITATION STYLE
Rathod, T., Shah, M., Shah, N., Raval, G., Bhavsar, M., & Ganesh, R. (2021). On Performance Enhancement of Molecular Dynamics Simulation Using HPC Systems. In Lecture Notes in Networks and Systems (Vol. 203 LNNS, pp. 1031–1044). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-981-16-0733-2_73
Mendeley helps you to discover research relevant for your work.