Packet-switching network architectures for very-high-speed service

Abstract

It is suggested that mesh-topology networks, deflection routing (DR), and backward learning (BL) are very attractive means for extending, to a geographical area, the services provided by LANs and MANs (local and metropolitan area networks). Unlike ring and bus topologies, mesh networks provide multiple paths between sources and destinations, which in turn offer very high connection reliability and traffic-handling capability. The DR technique switches, at locally clocked instants, the packets that are crossing the node. Address information and routing tables determine preferred outputs. If contentions occur, packets are deflected toward nonpreferred output links. Routing tables are maintained using the BL technique. Nodes following the described architecture and capable of switching packets among a few links (up to four) of equal speed (up to 1 Gb/s) can be implemented in hardware without a large technological effort. Since no packet buffering is provided in the network, control and maintenance procedures are extremely simplified with respect to classical packet-switching networks. The performance of the proposed architecture is evaluated, and its throughput efficiency is shown. In particular, it has the capability to respond to consistent topology changes without appreciable delay. >