The Polyp multiprocessor: Architecture and applications in nuclear physics Free

This article describes the architecture of the Heidelberg Polyp multiprocessor and its application in nuclear physics. The system is used for real‐time preprocessing and filtering of events that are read in from Fastbus front‐end electronics at a rate of ≤10⁸ parameters/s. The computing power is provided by a concurrent operation of many microprocessors (currently 30 Motorola 68000 CPUs with 16 kbytes of cache memory and 1 Mbyte of main memory each). They are interconnected by a multiple bus system, which allows adjustment of the bus bandwidth from 10 to 200 Mbytes/s. The processors and buses constitute pools that are managed by a decentralized hardware that permits making changes to the computing power and bus bandwidth without any software modifications. Another special feature is provided by the Syncbus that supports distributed scheduling by hardware. In addition, fault tolerance is achieved by forward and backward hardware error correction. Several systems have been runnning for more than 2 years, with one of them being currently upgraded to 48 processors (68020/68882 Motorola VME boards with 4 Mbytes each).