Parallel Data Refinement Layer of a Telescopic Approach for Extreme-scale Parallel Mesh Generation for CFD Applications

Abstract

In earlier work, we proposed a Telescopic approach which is a multi-layered approach for extreme-scale parallel mesh generation and adaptation. In this paper, we describe the Parallel Data Refinement (PDR) layer of the Telescopic approach. Namely focus on PDR's: (i) design and implementation and (ii) evaluation using TetGen, an open source mesh generation software, on shared memory machines. We outline lessons learned and future directions for revisiting the PDR layer and making adjustments in the implementation of the remaining layers of the Telescopic approach. Figure 1 Telescopic Approach to parallel mesh generation and adaptation and PDR layer in the middle.

On the End-User Productivity of a Pseudo-Constrained Parallel Data Refinement Method for the Advancing Front Local Reconnection Mesh Generation Software

“…There is still much to be accomplished in order to create a PDR.AFLR implementation. Based on past work and results [7], an unconstrained parallel data refinement implementation with AFLR is expected to be much faster and efficient than the pseudo-constrained PDR.AFLR. This will require a significant time investment, as many fundamental changes will need to be made to the source code.…”

“…Its methodology is theoretically proven to maintain stability and robustness for parallel isotropic Delaunay-based mesh generation and has been experimentally verified for isotropic meshes [4] [5] [6]. It is also designed to allow for the utilization of any sequential mesh generator while offering guarantees for the following five requirements for parallel mesh generation [7]: 1) Stability ensures that a mesh generated in parallel maintains a level of quality comparable to that of a sequentially generated mesh. This quality is defined in terms of the density and shape of the elements evaluated in the metric field, and the number of the elements (fewer is better for the same level of metric conformity).…”

“…Elements within a mesh may undergo refinement more than once when in parallel, so it is imperative that the sequential mesh generator satisfy this requirement. Previous work involving the integration of the mesh generator TetGen [8] with PDR shows that if the mesh generator fails to meet the reproducibility criterion in distributed memory, then the complexity of such state-of-the-art codes inhibits their modifications to a degree that their integration with parallel frameworks like PDR becomes impractical [7]. The original, sequential AFLR code was determined to be a suitable mesh generator to integrate with PDR, as it was tested and shown to maintain weak reproducibility.…”

A large‐scale parallel hybrid grid generation technique for realistic complex geometry

“…The other category of parallel mesh generation technology is completely data and algorithm parallel, rather than based on background mesh. The most representative work is parallel Delaunay mesh generation approach, which was first proposed by Chrisochoides et al And then it was expanded to multilayered parallel mesh generation . The multilayered mesh generation method has the capability of generating 18M elements per second on 256 cores while mathematically guarantee the quality and fidelity of the meshes.…”