Task-based Programming Models
In task-based models, an application is represented as a collection of tasks and a light-weight run-time system performs an analysis of data dependencies and schedules the tasks on appropriate resources. We will be using the Legion programming model and creating a task-based climate model from both the top down (replacing the top level coupler and driver layer) and the bottom up (replacing large components like the land, ocean and sea-ice with task-based versions). The figure above shows a dependency analysis generated by Legion for the current E3SM ocean model.
An important part of coupling multiple component models together is the remapping of fields from one mesh to another in a way that preserves important properties of the fields. Under the CANGA project, we are developing new remapping algorithms like:
Remapping techniques for irregular (non-convex) meshes like those for watersheds shown here,
Interpolation of velocity fields or other vector fields that preserve vector properties (e.g. divergence, curl)
Adaptive and time-dependent remapping to manage changing meshes (e.g. coastal inundation)
Mesh-free methods that are agnostic to how fields are positioned within a mesh (cell centers, edges, corners)
All of these new schemes will be tested in a new remapping test framework that we are developing to ensure results are both accurate and preserve the quantities we wish preserved.
Often, in a complex coupled model, each piece of the system is developed as a standalone component by experts in a particular sub-field. To create the final application, the components are coupled together by exchanging fields, often in an ad hoc manner based on computational or time-scale arguments. For example, many coupled climate models let the ocean, with its longer timescales, proceed concurrently and pass averaged fields from the previous coupling interval (hours or days). These choices, if not analyzed carefully, can introduce instability and error into the integrated system. CANGA scientists are performing such analyses of the coupled climate system and devising more accurate and stable methods for stepping the model forward for more accurate and robust projections of future change.