A load balancing technique for video signal processing on a multicomputer type DSP

To extract full performance from a multicomputer digital signal processing (DSP) system, the development of a viable load balancing technique is considered indispensable. The authors describe a task allocation and control methodology for image signal processing, such as is used in a low-bit rate (e.g. 64-kb/s) motion picture coding system. On a massive homogeneous multicomputer-type DSP system, the following approaches increase system performance substantially: (1) task scheduling with interframe prediction; (2) layered large-grain data flow; and (3) predefined local reassignment of processing. The effects of these various methods are examined in the context of a low bit rate motion picture codec with motion compensation and vector quantization application