ATM LAN interconnections with a cut-through nonblocking switch

Abstract

The paper presents a cut through ATM switch with shared buffers for LAN interconnections. The incoming ATM cells are not stored into shared buffers as long as the destination output port is available. In other words, an ATM cell is waiting at the input port. It will be passed through the switch using a cut through link, if the destination output port is available. On the other hand, it will be stored into the switch's shared buffers, if the destination output port is busy and the buffers are not full. Performance measures in terms of the end to end ATM cell delay, the switch cell delay, and the average number of cells queued in the shared buffers (defined as the required buffer size) are evaluated through simulation. In the simulation, it is assumed that the traffic load on each switch port is increased from 4 token rings (64 Mbps or 20% of port capacity) to 16 token rings (256 Mbps or 80% of port capacity). A simple model, based on the mean queue length of destination output ports, is used to analytically quantify the busy status of output ports. The simulation results point out that an ATM switch with cut through capability can significantly reduce the mean cell delay and the required buffer size, even when the probability of cells that can be cut through is very small. >