A general-purpose multi-micro-system with high fault-tolerance and unlimited system-capacity

Abstract The Heidelberg Multi-Processor-System is described. It will be used for online-data reduction, trigger processing, image processing, etc. The system consists of typically up to 300 processor-modules and several global memory modules. They are interconnected by a multi-common-bus for data transfer and a multiple synchronization bus for processor/task-scheduling. Hardware and software have been designed to be free of bottle-necks: the number of modules and the bus capacity are not limited. An increased system size does not lead to increased memory or bus interferences. The system has also been designed to be fault-tolerant: hardware and software are organized totally decentralized; most temporary and permanent memory and bus transfer faults are corrected by hardware. Defective modules or busses are disabled by software; the system continues its operation with almost the same speed.