Performance Improvements of a Parallel Multithreading Self-gravity Algorithm
- 551 Downloads
This article presents the application of performance optimization techniques to improve the computational efficiency of a parallel multithreading algorithm for self-gravity calculation on agglomerates. The studied algorithm applies the Discrete Element Method to simulate an ensemble of interacting particles under several contact and body forces. Based on the time scales of the process involved in the problem, we used a computation algorithm that speed up the self-gravity calculation based on defining a mesh over the simulated space. Specific performance improvements are presented, including the update of the occupied regions of the space, profiling and reimplementation of the most time consuming routines. Results indicate that the proposed implementation scale appropriately (almost-linear behavior) with the number of computational resources and the number of particles. The proposed improvements allow accelerating up to 50\(\times \) the execution times over the previous version of the self-gravity algorithm in the studied scenarios.
The work of Néstor Rocchetti, Sergio Nesmachnow, and Gonzalo Tancredi has been partly supported by CSIC, ANII, and PEDECIBA (Uruguay).
- 7.Abe, S., Altinay, C., Boros, V., Hancock, W., Latham, S., Mora, P., Place, D., Petterson, W., Wang, Y., Weatherley, D.: ESyS-Particle: HPC Discrete Element Modeling Software. Open Software License version 3 (2009)Google Scholar
- 14.Weatherley, D., Boros, V., Hancock, W., Abe, S.: Scaling benchmark of ESyS-particle for elastic wave propagation simulations. In: IEEE Sixth International Conference on e-Science, pp. 277–283. IEEE (2010)Google Scholar
- 15.Intel\(\textregistered \) vtune™ amplifier 2017 (2006). https://software.intel.com/en-us/intel-vtune-amplifier-xe. Accessed July 2017
- 16.Nesmachnow, S.: Computación científica de alto desempeño en la Facultad de Ingeniería, Universidad de la República. Revista de la Asociación de Ingenieros del Uruguay 61(1), 12–15 (2010)Google Scholar