Abstract: Data sets of sequential data strings at a particular point in time are assigned to processing elements of a multiprocessor system at high speed and processing is executed efficiently in parallel. A mechanism for preventing allocation due to all-read to a cache block which has not yet been referred from another cache is added by providing a reference bit with respect to cache blocks. The reference bit becomes 0 when new data is read to the cache block and becomes 1 when the cache block is referenced by a CPU. In the case of reading data from another cache, if a block corresponding thereto can be replaced without the need to be written back to the other cache and the reference bit is 1, data is fetched to the shared cache memory.

Abstract: A multiprocessor memory managing system and method make it possible for a series of instructions corresponding to a data set, which is sequentially renewed by a series of data, to be sequentially executed by a multiprocessor with small overhead. More specifically, the system and method provide a procedure dealing with a data set which is sequentially renewed by a series of data having a sequence, in which snap shots of the data set at arbitrary points of time are assigned to element processors of a multiprocessor at a high speed to execute instructions corresponding to the snap shots in an efficient and parallel fashion. The apparatus includes a master memory and first and second slave memories for storing renewable data. The first and second slave memories are alternately locked and read for snapshots of the renewable data.