Efficient Intersection of Distributed Meshes for use in the Community Earth System Model

Monday, May 12, 2014 - 07:00
Add to Calendar

[POSTER ABSTRACT] Computing intersections between two meshes covering the same domain is a problem appearing in many scientific applications, for example coupling between two different simulations codes, regridding, or in Arbitrary Lagrangian-Eulerian (ALE) type methods for computational fluid dynamics. For large meshes distributed across multiple processors, efficient and scalable algorithms are critical. One application is advection / transport in climate codes. The ALE method was developed in an attempt to combine the advantages of the Eulerian and the Lagrangian approaches by letting the mesh move in a prescribed manner as an extra independent degree of freedom. A popular choice of prescribed mesh movement is to run in Lagrangian mode for one time-step and then interpolate back to the static and regular (Eulerian) mesh. This interpolation step involves intersecting the Lagrangian mesh (also called departure mesh) and Eulerian mesh, and is known as semi-Lagrangian method in the meteorological literature. Here we present a linear complexity intersection algorithm that uses an advancing front approach when overlapping mesh sets are distributed across multiple processors. It is implemented using the MOAB (A Mesh-Oriented datABase) library. Preliminary performance results are shown on a leadership class machine. In the next steps of this work, we will leverage this method in the development of new advection schemes for both CAM-SE and MPAS-Ocean in the CESM. This work is a collaboration between computational scientists in the ACES4BGC Scientific Application Partnership project, computational mathematicians in the FASTMath Institute, and performance engineers in the SUPER Institute, all in SciDAC-3.

Presentation file(s):