Parallel Lattice Boltzmann Method with Blocked Partitioning

Abstract

This paper presents and discusses a blocked parallel implementation of bi-and three-dimensional versions of the Lattice Boltzmann Method. This method is used to represent and simulate fluid flows following a mesoscopic approach. Most traditional parallel implementations use simple data distribution strategies to parallelize the operations on the regular fluid data set. However, it is well known that block partitioning is usually better. Such a parallel implementation is discussed and its communication cost is established. Fluid flows simulations crossing a cavity have also been used as a real-world case study to evaluate our implementation. The presented results with our blocked implementation achieve a performance up to 31% better than non-blocked versions, for some data distributions. Thus, this work shows that blocked, parallel implementations can be efficiently used to reduce the parallel execution time of the method.

Scalable Parallel Algorithm of Multiple-Relaxation-Time Lattice Boltzmann Method with Large Eddy Simulation on Multi-GPUs

“…Owing to the local nature of LBM, it is a better choice to adopt domain decomposition method that partitions the flow domain into subdomain. As for three-dimensional flow simulation, there are 1D, 2D, and 3D (see Figure 4) domain decomposition methods [20], in which the fluid domain can be divided along one, two, or three directions. It can be seen from [20,25] that choosing the 3D one is the best option when running on cluster.…”

“…As for three-dimensional flow simulation, there are 1D, 2D, and 3D (see Figure 4) domain decomposition methods [20], in which the fluid domain can be divided along one, two, or three directions. It can be seen from [20,25] that choosing the 3D one is the best option when running on cluster. In the 3D one, each MPI process deals with a part of computational domain, and the process can be denoted as a triple ( , , ) according to the MPI process rank .…”

“…Wu and Shao [19] simulated the lid-driven cavity flow using MRT-LBM compared with SRT-LBM by parallel implementation. Schepke et al [20] presented blocked parallel implementation of D2Q9 and D3Q19 SRT-LBM; they divided computational domain into subdomains to reduce the MPI communication time. Since GPU 2 Scientific Programming has many computational units and the localized communication mode of each lattice well matches the data-parallel characteristic of GPU, Tölke [21] used shared memory to propagate parts of particle distribution functions (PDFs), avoiding costly misaligned access to the global memory on a single GPU, and Habich et al [22] extended this strategy from D2Q9 to D3Q19 LBM.…”

Multi-GPU thermal lattice Boltzmann simulations using OpenACC and MPI

“…In mono-dimensional partitioning, the computational domain is decomposed into several slices and each slice is allocated to a single GPU. An alternative strategy for domain decomposition is block partitioning, which is to decompose the domain in more than one dimension [45,46]. Figure 5 illustrates the block partitioning and data exchange for the 2D domain.…”

A scalable parallel unstructured finite volume lattice Boltzmann method for three‐dimensional incompressible flow simulations

“…Therefore, parallel strategies have to be considered to use the supercomputers. A number of works on the parallel LBM have been carried out on multi-CPUs 21,22 and multi-GPUs. [23][24][25][26][27][28] However, all of them were conducted on uniform Cartesian grids.…”