作者: R. Käppeli , S. C. Whitehouse , S. Scheidegger , U.-L. Pen , M. Liebendörfer
DOI: 10.1088/0067-0049/195/2/20
关键词: Domain (software engineering) 、 Cartesian coordinate system 、 Shared memory 、 Physics 、 Distributed memory 、 Domain decomposition methods 、 Code (cryptography) 、 Dissipation 、 Computational science 、 Theoretical physics 、 Computer cluster 、 Space and Planetary Science 、 Astronomy and Astrophysics
摘要: fish is a fast and simple ideal magneto-hydrodynamics code that scales to∼ 10 000 processes for Cartesian computational domain of∼ 1000 3 cells. The simplicity offish has been achieved by the rigorous application of operator splitting technique, while second order acc uracy maintained symmetric ordering operators. Between directional sweeps, three-dimensional data rotated in memory so sweep always performed cache-effi cient way along direction contiguous memory. Hence, only requires one-dimensional description conservation equations to be solved. This approach also enable an elegant novel parallelisation tha t based on persistent communications with MPI cubic decomposition machines distributed scheme then combined additional OpenMP different sweeps can take advantage clusters shared We document detailed implementation TVD advection flux reconstruction. magnetic fields are evolved cons trained transport scheme. show subtraction estimate hydrostatic gradien from total gradients significantly reduce dissipation simulations gravitationally bound objects. Through its effi ciency,fish as well-suited hydrodynamics classes large-sc ale astrophysical high-performance computer clusters. In preparation release public version, we demonstrate performance suite astrophysically orientated test cases.