Preprints
https://doi.org/10.5194/gmd-2021-296
https://doi.org/10.5194/gmd-2021-296
Submitted as: development and technical paper
27 Sep 2021
Submitted as: development and technical paper | 27 Sep 2021
Status: a revised version of this preprint is currently under review for the journal GMD.

Islet: Interpolation semi-Lagrangian element-based transport

Andrew M. Bradley, Peter A. Bosler, and Oksana Guba Andrew M. Bradley et al.
  • Sandia National Laboratories, Albuquerque, New Mexico

Abstract. Advection of trace species (tracers), also called tracer transport, in models of the atmosphere and other physical domains is an important and potentially computationally expensive part of a model's dynamical core (dycore). Semi-Lagrangian (SL) advection methods are efficient because they permit a time step much larger than the advective stability limit for explicit Eulerian methods. Thus, to reduce the computational expense of tracer transport, dycores often use SL methods to advect passive tracers. The class of interpolation semi-Lagrangian (ISL) methods contains potentially extremely efficient SL methods. We describe a set of ISL bases for element-based transport, such as for use with atmosphere models discretized using the spectral element (SE) method. An ISL method that uses the natural polynomial interpolant on Gauss-Legendre-Lobatto (GLL) SE nodes of degree at least three is unstable on the test problem of periodic translational flow on a uniform element grid. We derive new alternative bases of up to order of accuracy nine that are stable on this test problem; we call these the Islet bases. Then we describe an atmosphere tracer transport method, the Islet method, that uses three grids that share an element grid: a dynamics grid supporting, for example, the GLL basis of degree three; a physics grid with a configurable number of finite-volume subcells per element; and a tracer grid supporting use of our Islet bases, with particular basis again configurable. This method provides extremely accurate tracer transport and excellent diagnostic values in a number of validation problems. We conclude with performance results that use up to 27,600 NVIDIA V100 GPUs on the Summit supercomputer.

Andrew M. Bradley et al.

Status: final response (author comments only)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
  • CEC1: 'Comment on gmd-2021-296', Juan Antonio Añel, 25 Oct 2021
  • RC1: 'Comment on gmd-2021-296', Anonymous Referee #1, 29 Nov 2021
  • RC2: 'Comment on gmd-2021-296', Anonymous Referee #2, 04 Dec 2021
  • EC1: 'Comment on gmd-2021-296', James Kelly, 04 Dec 2021
  • AC5: 'Comment on gmd-2021-296', Andrew Bradley, 15 Dec 2021

Andrew M. Bradley et al.

Data sets

SL GPU performance data on Summit Mar 2021 Andrew M. Bradley https://github.com/E3SM-Project/perf-data/tree/main/nhxx-sl-summit-mar2021

Model code and software

Islet methods Andrew M. Bradley, Peter A. Bosler, Oksana Guba https://github.com/E3SM-Project/COMPOSE/releases/tag/v1.1.1

Performance-portable implementation in HOMME dycore Andrew M. Bradley, Peter A. Bosler, Oksana Guba https://github.com/ambrad/E3SM/releases/tag/islet-2d-paper-summit-sl-gpu-timings

Andrew M. Bradley et al.

Viewed

Total article views: 1,073 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
849 195 29 1,073 12 8
  • HTML: 849
  • PDF: 195
  • XML: 29
  • Total: 1,073
  • BibTeX: 12
  • EndNote: 8
Views and downloads (calculated since 27 Sep 2021)
Cumulative views and downloads (calculated since 27 Sep 2021)

Viewed (geographical distribution)

Total article views: 1,014 (including HTML, PDF, and XML) Thereof 1,014 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 20 May 2022
Download
Short summary
Tracer transport in atmosphere models can be computationally expensive. We describe a flexible and efficient interpolation semi-Lagrangian method, Islet. It permits using up to three grids that share an element grid: a dynamics grid for computing quantities such as the wind velocity; a physics parameterizations grid; and a tracer grid. Islet performs well on a number of validation problems and achieves high performance on up to 27,600 NVIDIA V100 GPUs on the Summit supercomputer.