Abstract

Multiprocessor interconnection networks are proposed which have the characteristic that each processor or memory module can connect to a group of its near neighbors and adjacent directly connected groups include some of the same processors or memories. These overlapping connectivity networks are attractive especially for a very large number of processors which cannot be provided with full connectivity but need to operate by communication between processors. Applications for overlapping connectivity networks include neural computers and dataflow computers. Bandwidth formulas are derived using a probabilistic approach, including when intermediate processors are used in the interconnection path. A general cell design is presented capable of representing various overlapping connectivity networks. >

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call