Optical bus based fully shuffled trees for massively parallel systems

Summary form only given. We present an optical implementation for fully shuffled trees (FSTs). The FSTs are hierarchical interconnection networks for parallel and massively parallel systems. They have the advantages of high fault tolerance, scalability, and adaptability to any basic processor element (PE) clusters. The optical bus based fully shuffled trees´ (OBFSTs) implementation is based on inverting the FST graph and the use of optical buses to interconnect communication processors at different levels of the tree. In comparison with other optical based interconnection networks, the OBFST has lower end-to-end delays and a lower number of i/o ports or optical connections per processor than the 1-dimensional, 2-dimensional optical arrays or the linear array of reconfigurable pipelined bus (LARPB). Also, each OBFST optical bus may have a constant number of processors irrespective of the system size.