Cores

Each page lists the activities within the different cores for metrics-driven extreme scale parallel anisotropic mesh generation

Introduction

Our long term goal is to achieve extreme-scale adaptive CFD simulations on the complex, heterogeneous High Performance Computing (HPC) platforms. To achieve this goal, we proposed in a telescopic approach (see Figure 1) [1] [2]. The telescopic approach is critical in leveraging the concurrency that exists at multiple levels in anisotropic and adaptive simulations. At the chip and node levels, the telescopic approach deploys a Parallel Optimistic (PO) layer and Parallel Data Refinement (PDR) layer, respectively (see Sections below). In this paper, we focus on the implementation of PDR. In future efforts, the PDR layer will be implemented on top of the PO layer [3]. The PDR layer relies on theory presented for parallel mesh generation and adaptivity [4] [5] [6]. The requirements for parallel mesh generation and adaptivity are: Figure 1 Telescopic Approach to parallel mesh generation and adaptation and PDR layer in the middle. 1. Stability: the quality of the mesh generated in parallel must be comparable to that of a mesh generated sequentially. The quality is defined in terms of the shape of the elements (using a chosen location/error-dependent metric), and the number of the elements (fewer is better for the same shape constraint). 2. Robustness: the ability of the software to correctly and efficiently process any input data. Operator intervention into a massively parallel computation is not only highly expensive, but most likely infeasible due to the large number of concurrently processed sub-problems. 3. Code re-use: a modular design of the parallel software that builds upon previously designed sequential meshing code, such that it can be replaced and/or updated with a minimal effort. Code re-use is feasible only if the code satisfies the reproducibility criterion, identified for the first time in this project. However, the experience from this project indicated that the complexity of state-of-the-art codes inhibits their modifications to a degree that their integration with parallel frameworks like PDR becomes impractical. 4. Scalability: the ratio of the time taken by the best sequential implementation to the time taken by the parallel implementation. The speedup is always limited by the inverse of the sequential fraction of the software, and therefore all non-trivial stages of the computation must be parallelized to leverage the current architectures with millions of cores. The experience from the PDR implementation suggests that the code-reuse criterion ought to be adjusted for several reasons: (i) complexity of industrial strength sequential mesh generation codes and (ii) their reproducibility properties.

Tightly-coupled approaches

This groups is experimented with two tightly-coupled approaches :

1. CDT3D
2. Parallel Optimistic Delaunay Mesh Generation for medical images, for details see [LINK]

Partially-coupled approaches

==== Data Decomposition ====

1. Parallel Data Refinement using TetGen
2. Parallel Data Refinement using AFLR
3. Parallel Data Refinement using Parallel Optimistic Delaunay Mesh Generation for medical images , for details see [LINK to Daming papers]

==== Domain Decomposition ====

LINK to 2d PCDM paper

extension to 3D is in progress.

Loosely-coupled approach

Loosely coupled using CDT3D [MiniRTS workshop slides]

Real-time Medical Image Computing

Real Time Medical Image Computing