Self-Adaptive Load Sharing for Distributed Image File Processing

Many computational tasks in image processing are dataseparable in the sense that the correctness of results is irrelevant to the order of computing individual pixels. By taking advantage of this property, we have implemented a distributed image file processing (DIFP) program on an Ethernet network and have illustrated substantial speed-up. A master/slave model for DIFP is proposed, which is based on a self-adaptive load sharing protocol -“dispatching by request”. Issues in design and implementation of DIFP systems including load sharing, data partitioning, job distribution, concurrent file access, and error handling have been addressed.