Hervé Le Berre : November 21, 2005

Parallelization of an atmospheric model and an introduction to physics in the (mathematical) algo world.

Abstract

This informal presentation will be devoted to the optimization and parallelization of the atmospheric chemistry model MOCAGE from Meteo-France. Two ways of parallelization are considered: Open MP on cluster of SMP, and MPI for distributed memory machines. Both approaches, while technically different, share the same goal: making MOCAGE run faster. The proposed parallel remedies involves a basic understanding of the underlying physics in MOCAGE, particularly in the domain of meteorology and oceanography fields. The computational modifications will consequently be briefly motivated by introducing the implied physical concepts.