Advanced frame recovery in switched connection inverse multiplexing for ATM

Switched connection inverse multiplexing for ATM (SCIMA) has been proposed as a low-overhead scheme for implementing an N×N switch with a port rate of kR using a kN×kN core switch (or k N×N switches) having a port rate of R. By splitting the traffic of a small number of virtual connections into subconnections over multiple low-capacity core switch ports, SCIMA achieves a performance level comparable with that of an equivalent high-capacity port switch. The scheme ensures that each split cell flow is re-assembled and delivered in the original order even in the presence of variable differential delays and highly non-uniform switching path usage. Furthermore, it can seamlessly recover from isolated cell losses within the core switch, as well as detect the loss of a certain number of consecutive cells belonging to the same subconnection. However, it is unable to restore, by itself, the re-assembly cell chain after two or more consecutive cell losses occur in a row on the same subconnection, and has to rely on other recovery means. In this paper, we describe a fault tolerance mechanism called advanced frame recovery, which applies the concept of framing in the asynchronous environment of SCIMA by inserting regular, albeit not necessarily periodic, checkpoints into the split cell flow. When the mechanism is invoked, it discards some of the successfully delivered cells and allows us to restart the chain at a subsequent checkpoint. The advanced frame recovery scheme is used when SCIMA detects consecutive cell losses within the same subconnection; it also serves as an additional measure of integrity checking during regular re-assembly to correct certain vulnerabilities of SCIMA. We present the motivation and implementation details of advanced frame recovery, together with a discussion of the trade-offs of such an open loop (as opposed to feedback driven) sequence control scheme