OpenCAL is a scientific software library specifically developed for the simulation of 2D and 3D complex dynamical systems on parallel computational devices. It is written in C/C++ and relies on OpenMP/OpenCL and MPI for parallel execution on multi-/many-core devices and clusters of computers, respectively. The library provides the Extended Cellular Automata paradigm as a high-level formalism for modeling complex systems on structured computational grids. As a consequence, it can be used as a parallel explicit solver of ordinary and partial differential equations through classical methods, including finite-difference and finite-volume. Here the latest version of the library is described, introducing the MPI infrastructure over the 3D OpenCL and 2D/3D OpenMP components. The implementation of a three-dimensional unsaturated flow model based on a direct discrete formulation of the Richards' equation is also shown, corresponding to a finite-difference scheme. Computational performances have been assessed on both a scientific workstation equipped with a dual Intel Xeon socket and three Nvidia GPUs, and a 16 nodes cluster with a fast interconnection network. The OpenCAL embedded quantization optimization is also discussed and exploited to drastically reduce computing time.
OpenCAL system extension and application to the three-dimensional Richards equation for unsaturated flow
Giordano A;
2020
Abstract
OpenCAL is a scientific software library specifically developed for the simulation of 2D and 3D complex dynamical systems on parallel computational devices. It is written in C/C++ and relies on OpenMP/OpenCL and MPI for parallel execution on multi-/many-core devices and clusters of computers, respectively. The library provides the Extended Cellular Automata paradigm as a high-level formalism for modeling complex systems on structured computational grids. As a consequence, it can be used as a parallel explicit solver of ordinary and partial differential equations through classical methods, including finite-difference and finite-volume. Here the latest version of the library is described, introducing the MPI infrastructure over the 3D OpenCL and 2D/3D OpenMP components. The implementation of a three-dimensional unsaturated flow model based on a direct discrete formulation of the Richards' equation is also shown, corresponding to a finite-difference scheme. Computational performances have been assessed on both a scientific workstation equipped with a dual Intel Xeon socket and three Nvidia GPUs, and a 16 nodes cluster with a fast interconnection network. The OpenCAL embedded quantization optimization is also discussed and exploited to drastically reduce computing time.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.