Keywords: smoothed particle hydrodynamics, particle flow, parallelization, domain decomposition, message passing interface, single instruction multiple data.

Numerical simulation using computers is widely used to solve complex physical problems involving fluids or solids in engineering. Meshfree methods, such as smoothed particle hydrodynamics (SPH), are very attractive numerical methods for simulating fluid flow because of their advantages over grid-based methods for problems dealing with complicated phenomena.

In the SPH method, a set of particles in the domain can be associated with one discrete physical object respectively and each particle possesses individual mass, momentum and other material properties. The particles are free to move and have an effect on the neighbouring particles within the support domain which can vary with time. These features of the SPH method contribute to the computational time and expense. Therefore, parallel computing for the SPH method is necessary to improve the numerical efficiency for problems involving large computational domains. In this case, the message passing library, message passing interface (MPI), is used for communication between processes to deliver data to each other.

In the case of the parallelization of the SPH method, the decomposition of the analysis space or operation is an important step. Morris et al. [1] and Wu and Tseng [2] presented a suitable method for domain decomposition in the SPH method. There are also several decomposition approaches based on the particle, domain and operation [3,4]. In this paper, the partition of the simulation domain is proposed for the continuity of subdomain boundary for the single instruction multiple data (SIMD) structure. For the domain decomposition used in the proposed method, the boundary conditions are composed of the interface particles instead of the ghost zone of the other side of the subdomain. In this case, the interface particle plays a role of filtering the particles near the boundary.

Using parallel computing, with the new proposed parallel method, the SPH method is more flexible for better performance. The running time of subdomains and the number of particle iterations are compared according to the number of decomposed domains and filtering step. Finally, the performance of the parallel computing of the SPH method is evaluated by the simulation of a fluid flow.

1

J.P. Morris, Y. Zhu, P.J. Fox, "Parallel Simulation of Pore-scale Flow through Porous Media", Computers and Geotechnics, 25, 227-246, 1999. doi:10.1016/S0266-352X(99)00026-9

2

J.S. Wu, K.C. Tseng, "Parallel DSMC Method using Dynamic Domain Decomposition", International Journal for Numerical Methods in Engineering, 63(1), 37-76, 2005. doi:10.1002/nme.1232

3

D.W. Holmes, J.R. Williams, P. Tilke, "A Framework for Parallel Computational Physics Algorithms on Multi-core: SPH in Parallel", Advances in Engineering Software, 42(11), 999-1008, 2011. doi:10.1016/j.advengsoft.2011.05.017

4

G.R. Liu, M.B. Liu, "Smoothed Particle Hydrodynamics: A Meshfree Particle Methods", World Scientific, 2003. doi:10.1142/9789812564405

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £50 +P&P)