Optimization of Load Distribution in Electronic Holography with a Parallel Computer.

Although electronic holography requires a large number of computations, we succeeded in generating holographic images, even moving images, using a parallel computer “CYBERFLOW” with 64 processors. Because of the computer's performance, however, the number of point sources in the image was limited. In this paper, we attempt to speed up the generation of images including more point sources, using a CYBERFLOW with 128 processors. Generation speed, however, does not increase in proportion to the number of processors with our conventional method of distributing the hologram among the processors, because this method results in differences between their loads. We propose an algorithm that distributes the hologram among the processors so that the difference between their loads can be reduced. Applying this algorithm to the 128 processors, we achieve a generation speed that is 2.76 times higher than that of the conventional method. We also propose a method for finding the most appropriate value of the parameter for the dynamic distribution.