Advanced parallelization strategies using hybrid MPI-CUDA octree DSMC method for modeling flow through porous media

Revathi Jambunathan, Deborah A. Levin

Research output: Contribution to journalArticlepeer-review

Abstract

The advantages of a linear space filling Morton Z-curve to represent an unbalanced three-dimensional octree structure for the Direct Simulation Monte Carlo method are assessed. The strategies to optimize and exploit the properties of the linearized tree using simple, binary computations are presented. Hybrid MPI-CUDA communications are invoked to facilitate the use of heterogeneous architectures for large-scale computations. Strong scaling studies have shown that the parallelization strategies implemented in this work results in 85% efficiency, and weak scaling studies show nearly 100% efficiency for a problem size with 0.34 billion particles and 1.5 million immersed body surface elements. Two types of problems, supersonic external flow over fractal-like immersed body and subsonic internal flow through a porous material are solved using the multi-GPU DSMC solver. The permeability of Morgan carbon felt material is calculated by modeling the diffusion of argon gas through the material and the calculated continuum permeability values match well with published data.

Original languageEnglish (US)
Pages (from-to)70-87
Number of pages18
JournalComputers and Fluids
Volume149
DOIs
StatePublished - Jun 13 2017

Keywords

  • DSMC
  • Forest of unbalanced linear octrees
  • Heterogeneous architectures
  • Large-scale simulations
  • MPI-CUDA
  • Morton-Z curves and Morton encoding
  • Permeability
  • Porous media

ASJC Scopus subject areas

  • General Computer Science
  • General Engineering

Fingerprint

Dive into the research topics of 'Advanced parallelization strategies using hybrid MPI-CUDA octree DSMC method for modeling flow through porous media'. Together they form a unique fingerprint.

Cite this