Toward the design of a novel hybrid parallel N-body method in scope of modern cloud architectures

摘要

A hybrid parallel self mesh-adaptive N-body method based on approximate inverses and multiprojection techniques is proposed. This method is a three-dimensional hybrid parallel mesh-type N-body scheme based on the solution of the Poisson equation in the physical space with boundary conditions obtained from multipole expansion formulas. In order to improve the accuracy of the solution, especially in shallow regions, a self mesh-adaptive scheme is used to create a hierarchy of independent smaller N-body problems. The parallelization of the scheme is based on a uniform partitioning of the bodies with respect to available computer nodes, and communications are required only for the computation of the density and potential distributions. The proposed scheme is suitable for large-scale galaxy simulations with millions of bodies on high-resolution meshes, for distributed HPC systems with multicore computer nodes. Moreover, large-scale galaxy simulations are performed on modern Cloud environments in order to examine the applicability and performance. Implementation issues concerning the proposed scheme are also discussed. The parallel performance and speedup of the hybrid parallel N-body method on HPC systems as well as on Cloud environments are presented and discussed.