Adaptive Mesh Refinement applications and requirements Vincent Pichon COOP 9 mars 2009 Outline Motivation Adaptive Mesh Refinement AMR based on component ULCM / Java Conclusion Motivation Solving equations on a discrete domain Spacing determines local error => accuracy of the solution Spacing determines number of calculations. Uniform mesh spacing fine enough is extremely costly => Adaptive Mesh Refinement Adaptive Mesh Refinement Start with a coarse mesh Locally refine only where needed Concentration of effort where it is needed Multilevel 0 Domain level 0 Adaptive Mesh Refinement Start with a coarse mesh Locally refine only where needed Concentration of effort where it is needed Multilevel 1 1 1 1 Adaptive Mesh Refinement 2 2 1 Locally refine only where needed Concentration of effort where it is needed Multilevel 2 2 2 2 2 2 1 Adaptive Mesh Refinement 2 Locally refine only where needed Concentration of effort where it is needed Multilevel 2 2 3 3 3 3 1 2 2 2 2 1 AMR based on component Each component has a maximum mesh size. Component Coarse mesh covering the whole domain Component Recursion Replace component by composite AMR based on component Increase Global size of mesh EDF R&D : Créer de la valeur et préparer l’avenir ULCM / Java Application ULCM / Java Reconfiguration of connections and creation of new instances ULCM / Java Compute service Conclusion Existing applications (DAGH, ENZO, CHOMBO, CLAWPACK...) Based on MPI Load balancing on static resources Component approach: Resource dynamicity Placement control