Year: 1995
Type: conference-abstract
Abstract: We have developed a gravity solver based on combining the particle-mesh (PM) method and TREE methods. It is designed for and has been implemented on parallel computer architectures. The new code can deal with tens of millions of particles on current computers, with the calculation done on a parallel super- computer or a group of workstations. Typically, the spatial resolution is enhanced by more than a factor of 20 over the pure PM code with mass resolution retained at nearly the PM level. This code runs much faster than a pure TREE code with the same number of particles and maintains almost the same resolution in high-density regions. Multiple time step integration has also been implemented with the code, with second-order time accuracy. The performance of the code has been checked in several kinds of parallel computer configurations, including IBM SP1, SGI Challenge, and a group of workstations, with the speedup of the parallel code on a 32 processor IBM SP2 supercomputer nearly linear (efficiency ≍ 80%) in the number of processors. The computation/communication ratio is also very high (∼50), which means the code spends 95% of its CPU time in computation.
DOI: 10.1086/192166
Author: Guohong Xu
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2043064688