Performance analysis of queueing models for multicast ATM switch architectures

A copy network followed by a point-to-point switching network is a design approach that has been traditionally used for multicast switch architectures. One of the inherent problems in this design is the overflow which occurs when the total number of copy requests exceeds the network size. A natural solution is to store the blocked packets in a buffer. This paper investigates the performance of the shared-memory input buffering architecture, and compared it with that of the dedicated-memory input buffers. Results from both analysis and simulation are provided, which show the shared-memory buffering scheme with better performance.<<ETX>>