A MPI Parallel DGTD Method with Local Time Stepping Technique

This paper proposed a MPI parallel algorithm of discontinuous Galerkin time domain method (DGTD) with local time stepping technique. In order to solve large multi-scale electromagnetic problems with large-scale CPU cores, two-layer parallel transmission and sparse matrix storage format are used to reduce the time of preprocessing. In time stepping procedure, the persistent non-blocking MPI communication method is adopted to overlap the communication and calculation to improve parallel efficiency. Numerical results are given to demonstrate good performance of the proposed method up to 30,000 cores.