Efficiency of parallel computations of gravitational forces by TreeCode method in N-body models

TL;DR

This paper evaluates the computational efficiency of parallel TreeCode algorithms for N-body gravitational simulations, demonstrating a 20% average performance increase when utilizing all logical threads on various processor architectures.

Contribution

It provides an analysis of parallel TreeCode performance on different architectures, highlighting the impact of thread utilization on efficiency in large-scale N-body simulations.

Findings

Parallel TreeCode improves computation speed for large N-body models.

Using all logical threads increases efficiency by about 20%.

Performance gains depend on processor architecture and thread management.

Abstract

Modeling of collisionless galactic systems is based on the N-body model, which requires large computational resources due to the long-range nature of gravitational forces. The most common method for calculating gravity is the TreeCode algorithm, which provides a faster calculation of the force compared to the direct summation of contributions from all particles for N-body simulation. An analysis of the computational efficiency is performed for models with the number of particles up to $10^{8}$. We considered several processors with different architectures in order to determine the performance of parallel simulations based on the OpenMP standard. An analysis of the use of extra threads in addition to physical cores shows an increase in simulation performance only when all logical threads are loaded, which doubles the total number of threads. This gives an increase in the efficiency of parallel computing by 20 percent on average.