Parallel load‐balanced simulation for short‐range interaction particle methods with hierarchical particle grouping based on orthogonal recursive bisection

Abstract

SUMMARYWe describe an efficient load-balancing algorithm for parallel simulations of particle-based discretization methods such as the discrete element method or smoothed particle hydrodynamics. Our approach is based on an orthogonal recursive bisection of the simulation domain that is the basis for recursive particle grouping and assignment of particle groups to the parallel processors. Particle grouping is carried out based on sampled discrete particle distribution functions. For interaction detection and computation, which is the core part of particle simulations, we employ a hierarchical pruning algorithm for an efficient exclusion of non-interacting particles via the detection of non-overlapping bounding boxes. Load balancing is based on a hierarchical PI-controller approach, where the differences of processor per time step waiting times serve as controller input.

The comparison of two domain repartitioning methods used for parallel discrete element computations of the hopper discharge

“…In our implementation, each processor computes the forces for particles located in the processor subdomain, while extra particles, residing in ghost regions, are stored only for data required for these computations. An alternative solution based on the symmetry of contact forces would be to calculate the contact forces for the ghost particles and to send the obtained values to the neighbouring subdomains [25]. The current time step is finished by the condition, which redirects the program flow to the end of computations if the end of the investigated time interval is reached.…”

“…A higher efficiency of load balancing is obtained, adjusting the balance among the hierarchically grouped domains by compressing and stretching the cells in a group. Fleissner and Eberhard [25] applied an orthogonal recursive bisection of the simulation domain for recursive particle grouping and assignment to parallel processors. The parallel speed-up of 10.2 was achieved by using 16 nodes and simulating 1.0 Â 10 6 of particles.…”

Massively parallel rigid body dynamics simulations

“…Fleissner et al [14] focus on parallel simulation of particle-based discretization methods, such as the discrete element method or smoothed particle hydrodynamics, instead of rigid body dynamics. Although their approach offers the option to run large-scale particle simulations, they are only considering point masses instead of fully resolved rigid bodies.…”

Massively parallel granular flow simulations with non-spherical particles

“…For instance, Fleissner et al [4] focus on load balancing strategies for the massively parallel simulation of particle-based discretization methods, such as the discrete element method or smoothed particle hydrodynamics. However, although their approach offers the option to run largescale particle simulations, they are only considering point masses instead of fully resolved rigid bodies.…”

“…Several numerical simulation approaches, mainly based on the representation of granular particles as point masses, have been used for the simulation of granular media, such as molecular dynamics [2,3], Monte Carlo simulations, cellular automata [3], the discrete element method, or smoothed particle hydrodynamics [4].…”