The International Conference for High Performance Computing, Networking, Storage and Analysis
Greater than 10x Acceleration of Fusion Plasma Edge Simulations Using the Parareal Algorithm.
Authors: Debasmita Samaddar (Culham Centre for Fusion Energy/UK Atomic Energy Authority), David P. Coster (Max Planck Institute of Plasma Physics), Xavier Bonnin (National Center for Scientific Research), Christoph Bergmeister (University of Innsbruck), Eva Havlickova (Culham Centre for Fusion Energy/UK Atomic Energy Authority), Wael R. Elwasif (Oak Ridge National Laboratory), Lee A. Berry (Oak Ridge National Laboratory), Donald B. Batchelor (Oak Ridge National Laboratory)
Abstract: Simulations involving the edge of magnetically confined, fusion plasma are an extremely challenging task. The physics in this regime is particularly complex due to the presence of neutrals and the device wall. These simulations are extremely computationally intensive but are key to rapidly achieving thermonuclear breakeven on ITER-like machines. Space parallelization has hitherto been the most common approach to the best of our knowledge . Parallelizing time adds a new dimension to the parallelization thus significantly increasing computational speedup and resource utilization. This poster describes the application of the parareal algorithm to the edge-plasma code package – SOLPS. The algorithm requires a coarse (G) and a fine (F) solver – and optimizing G is a challenge. This work explores two approaches for G leading to computational gain >10 which is significant for simulations of this kind. It is also shown that an event-based approach to the algorithm greatly enhances performance.