An Efficient Parallel BP for Large Scale Training-sets and its Implementation on the Multicomputer System