Event Detail

Event Type: 
M.Sc. Presentation
Tuesday, October 26, 2021 - 10:00 to 12:00
Zoom - Email nikki.sullivan@oregonstate.edu for Zoom log in details

The Model for Prediction Across Scales-Ocean (MPAS-O) is an ocean model built on an unstructured mesh framework that allows for variable spatial resolution. A major hindrance of running MPAS-O efficiently when using traditional time-stepping schemes is that the largest time-step that can be selected is bounded above in terms of the size of the smallest cell on the mesh due to the CFL condition. To address this problem, Hoang et al. developed local time-stepping (LTS) schemes that allow for the selection of different sized time-steps depending on the local resolution of the mesh. These LTS schemes were then implemented in the shallow water core of MPAS-O by Capodaglio and Petersen. LTS schemes promise considerable speedups (compared to traditional time-stepping methods) in global ocean simulations. In this talk, we present the results of three experiments relating to the performance of LTS in the MPAS framework. In the first two, we compare the performance of a particular LTS scheme to that of the classical fourth-order Runge-Kutta method as we vary parameters relating to the resolution of the spatial mesh. In the third, we investigate the performance scaling of LTS across number of processors. These experiments demonstrate that LTS can provide significant speedups versus traditional global time-stepping methods.