Generalized inverse multiplexing of switched ATM connections

Inverse multiplexing for ATM (IMA) has been standardized by the ATM Forum to provide a high-capacity logical link by grouping several lower-capacity physical links. IMA is intrinsically point-to-point, assumes congestion-free links, and only tolerates relatively constant differential delays between the links. In this paper, we generalize the notion of inverse multiplexing by introducing switching and buffering within the inverse multiplexed segment. The new scheme, which we call Switched Connection Inverse Multiplexing for ATM (SCIMA) allows to implement an N/spl times/N switch with ports operating at rate kR by using a kN/spl times/kN core ATM switch with ports of rate R. SCIMA executes demultiplexing on a selected set of virtual connections, dynamically distributes their traffic onto a group of switching paths, and performs egress re-assembly using an asynchronous method for differential delay compensation. The scheme is simple to implement, has low overhead, is robust in presence of cell loss within the switch, handles congestion, and tolerates wide variations in differential delays.