https://hgpu.org/?p=8250
Overlapping computation and communication of three-dimensional FDTD on a GPU cluster