Starting from categories of the known computer-networks-area like message-handling, bus, protocol etc. may lead to a dead-end with respect to multiprocessor-design. Those terms from the world of networks may persuade computer architects to choose solutions which are not adequate to the requirements of efficient and fault-tolerant operation.
Topological investigations including technological considerations will result in structures which differ contrarily from the traditional monoprocessor as well as from computer networks. The macro-dataflow-concept [11] for instance can ensure in the framework of the usual storage-access a high rate of efficiency. Also, system programming and application programming will not deviate in this case too much from the traditional patterns. Beyond it the approaches for fault-tolerant operation become then simple and effective.
Experiences with multiprocessors, which are accordingly designed are discussed.