Parallel Efficiency of an Adaptive, Dynamically Balanced Flow Solver

Gepner, S. W.; Majewski, J.; Rokicki, Jacek

doi:10.1007/978-3-642-55195-6_51

Cited by 1 publication

(5 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The lines marked with ( ) correspond to the results obtained for the original, unadapted mesh, resulting in a problem containing 8.0•10 5 DOFs [16]. For this particular configuration superlinear speed-up has been achieved (smaller tasks better fit into the processor cache memory).…”

Section: The Load Balancingmentioning

confidence: 96%

“…Even if the load is balanced at the beginning of the simulation, the dynamic changes caused by the adaptive refinement will significantly lower the efficiency [10,16]. This makes direct application of adaptivity unfeasible for the automated parallel simulations.…”

Section: The Load Balancingmentioning

confidence: 99%

“…In this work we make use of the approach described earlier in [12] and [16]. The impact of this particular DLB implementation on the load balance indicator β, and in consequence on the parallel performance is summarized below.…”

Section: The Load Balancingmentioning

confidence: 99%

“…To this end, a range of partitioning methods, based on graph partitioning algorithms, have been developed [5][6][7][8]. Since adaptivity delivers a dynamic change to the distribution of mesh entities among processors a dynamic load balancing strategy must be used to maintain high performance of parallelization [9][10][11][12][13][14][15][16].…”

Section: Introductionmentioning

confidence: 99%

“…Majewski, and J. Rokicki Due to the template based refinement, there is no need to calculate the metric field directly, instead an adaptive indicator [12,53] is used to trigger edge splits. The dynamic load balancing strategy (DLB) [11,16] is applied to maximize the parallel effectiveness of calculations. This paper is organized as follows.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Parallel anisotropic mesh refinement with dynamic load balancing for transonic flow simulations

Gepner

Majewski

Rokicki

2017

Bulletin of the Polish Academy of Sciences Technical Sciences

View full text Add to dashboard Cite

Abstract. The present paper discusses an effective adaptive methods suited for use in parallel environment. An in-house, parallel flow solver based on the residual distribution method is used for the solution of flow problems. Simulation is parallelized based on the domain decomposition approach. Adaptive changes to the mesh are achieved by two distinctive techniques. Mesh refinement is performed by dividing element edges and a subsequent application of pre defined splitting templates. Mesh regularization and derefinement is achieved through topology conserving node movement (r-adaptivity). Parallel implementations of an adaptive use the dynamic load balancing technique.Key words: parallel CFD, adaptation, dynamic load balancing, mesh refinement, parallel efficiency.Parallel anisotropic mesh refinement with dynamic load balancing for transonic flow simulations [28][29][30][31][32]. In general, this approach yields high quality meshes for arbitrarily complex geometries [4,[33][34][35][36]. Nevertheless this approach is difficult to implement in parallel simulations because of the global nature of the remeshing operation.An alternative approach is to apply modifications to the computational mesh only in the regions of interest. Either by local mesh modification operators (edge splits, edge collapses and edge swaps) [37][38][39][40] or element bisection (e.g., based on template refinement) [10,[41][42][43][44][45][46][47]. Locality of these methods makes them especially attractive for application in parallel computing, as the necessary communication volume and frequency become limited, boosting parallel performance.Another possibility is brought by the r-adaptive methods, in which grid nodes are allowed to move without changing the topology of the mesh. This technique is used either to augment local refinement methods [48,49], or as an adaptive tool in itself [50][51][52]. Application of r-adaptive algorithms might be beneficial in parallel computations, as neither the grid topology nor the load balance become affected.Although adaptive methods are well recognised in the academic community, the broader application of adaptation for industrial problems has been limited. This paper presents an approach towards fully parallel and automatic mesh adaptation method that can be used for 3D industrial cases. The h-r adaptive method is developed, suitable for parallel simulation of transonic flows. Adaptivity is accomplished by anisotropic template-based element bisection method. The coarsening step is carried out via an elastic analogy node movement. Anisotropic mesh modifications are driven by a Hessian based approach.

show abstract

Section: The Load Balancingmentioning

confidence: 96%

Section: The Load Balancingmentioning

confidence: 99%