The availability of exascale computers with >106 cores will enable large-scale unsteady simulations across a variety of multi-physics applications, modelled using complex, evolving geometrical domains.
Many of the grand challenge problems that will exploit the exascale will require very large meshes with 109 cells or more, together with cells that may differ in size by many orders of magnitude to faithfully resolve the underlying physics at the appropriate scales.
Since the size of exascale meshes means that “serial” generation of an initial mesh is no longer viable, distributed parallelisation of the whole workflow is necessary. Current limitations mean that, at present, most meshes are generated in a multi-threaded, shared memory environment, rather than upon a distributed memory parallel system.
To enable exascale simulations, issues that must be addressed include improved geometry handling, mesh adaptation and optimisation, intelligent meshing and automation and robustness, all within a large, distributed environment that lies outside of current capabilities.
Solutions to some of these issues that this project is investigating are:
- Distributed representation of the underlying CAD system
- Parallelisation of the mesh generation process
- Automatic mesh sizing and adaptivity
- Development of fault-tolerant processes within workflows that will mitigate hardware failure impact
- High-order mesh generation
The aim of the Exascale Mesh Network (ELEMENT) project is to ensure that meshes of sufficient quality and resolution to represent exascale problems can be generated efficiently using distributed memory systems in order to prepare the way for highly scalable solutions that will exploit extreme levels of parallelism.
The central goal is to address the significant bottleneck that pre-processing/mesh generation poses in the development of adaptive numerical solvers for the exascale.
ELEMENT’s objectives are therefore:
- To build a community around meshing best practice by establishing a collaborative network.
- To undertake proof of concept studies, with prototype implementations of two target challenges.
- To publish a Vision Paper and strategic research agenda covering the full meshing workflow, from mesh generation to adaptation, partitioning and visualisation.