Infrastructure, parallel component model and initial results for coupled core-edge fusion simulationsJohn Cary, Tech-X Corp. |
The Framework Application for Core-Edge Transport Simulations (FACETS) is developing a parallel, scalable infrastructure allowing for multi-physics simulations that couple core transport, MHD equilibrium, edge transport and wall models. This will provide integrated modeling capability for simulating burning plasmas in ITER-like devices. The challenges facing FACETS include time scales varying over many orders of magnitudes from plasma-wall interaction (~1 micro-second), turbulent mixing (~1 ms), to the slow evolution of averaged profiles (~1 s), all the way to discharge durations exceeding 100 seconds.
The infrastructure is written in C++ and uses an abstract interface design to avoid explicit dependencies on particular implementations. For example, the I/O is done through an abstract API that is independent of the actual underlying library used. A concrete implementation using HDF5 has been implemented, but other mechanisms (e.g., netCDF) and/or data organizations could also be used. The use of C++ templates allows us to provide a uniform interface for a variety of data types and space dimensions and minimizes coding. All code adheres to a strict layering principle that prevents circular dependencies.
A component/coupler system allows coupled multi-physics simulations. The components form a n-ary tree with individual physics updaters as leaf nodes. Each component can be initialized on a separate set of processing elements and communication between nodes is handled by parent components using a client-server model. This allows development and testing of each individual component separately, and then coupling the component into the system as described by an XML-like input file.
A number of simplified models are being developed. As well wrapping existing models is being performed for coupling into the system. Initial results of coupling core and edge modules to perform parallel coupled simulations and scaling study results will also be presented.