Integrating the BSP400 neurocomputer with the MetaNet network environment

The authors describe the integration of hardware and software components into a single neural network environment. The building blocks described are the BSP400, a massively parallel neurocomputer, and MetaNet, a general neural network design tool. This research is a feasibility study for future developments in large-scale integration of computational resources. The objective is to obtain a neural simulation environment that supports easy development and efficient parallel calculation. Integration takes place using device drivers. Apart from handling communications between the components the primary task of the device driver is the mapping of a neural network definition onto the processors of the parallel hardware such as the BSP400. This mapping problem is caused by hardware constraints, such as maximum amount of connections and the topology of the processor network. The ultimate goal is an environment in which the user can program a single `ideal virtual´ machine, which in reality is assembled from many different computational resources