| 摘要 |
An ATM-switch is built in tree stages, each consisting of a plurality of switch cores. Each switch core in the first stage is duplicated to form a redundant pair of planes operating independently of each other. A cell from an input unit is thus sent to both planes in such a pair. All the switch cores in this stage work in a distributed mode, distributing a cell incoming to one of its input ports to a random one of its output ports. A plane selection is made after each such pair of switch cores. In that way the cells are also distributed at random to all switch cores in the middle stage. This stage acts in a load-sharing way, in which the switch cores act in the conventional switching way and thus the switch cores are not duplicated. A rather high number of switch cores should be used in this stage since, in the case of a faulty switch core in the middle stage, the remaining switch cores have to carry all the traffic of cells. The switch cores of the second stage will direct the cells to the coffect switch core of the third stage. The third stage has like the first stage duplicated planes. This gives a switch in which the switch cores are efficiently used. It is also reliable and robust since only a fraction of the traffic is at most affected by a fault in a switch core and it is tolerable to multiple faults, as long as they affect different switch cores. |
