Local Time Stepping and Load Balance of DGTD on High-Performance Computing Architectures

Extremely small meshes of system-in-package severely limit the integration-time step of simulations of time-domain electromagnetic problems. Here, we redesigned a parallel multilevel local time stepping (LTS) algorithms as well as load balance scheme to make our code capable of running on large clusters with high parallel efficiency. Besides, communication cost can be reduced by adjusting the update order of mesh levels. Numerical experiments are given to validate our algorithms.