Making multicast parallel packet switch practical

Abstract

In recent years, parallel packet switch (PPS) has become a central topic of discussions and researches. Because PPS is a promisingly practical scheme for us to build high-capacity packet switches that can support individual line-rates in excess of the speeds of available electronic memory and precisely emulate an FCFS OQ switch performance with a small speedup. But, most of the promising results presented so far assume the input traffic pattern is unicast-only. When the input traffic pattern is multicast traffic, however, the theoretic speedup requirement of an TV-port PPS increases to 2/spl radic/ + 1. Clearly, the resulting explosive increase of speedup makes these traditional multicast PPS schemes, which are just relied on speedup, impractical for any meaningful N. Instead of increasing speedup as traditional PPS schemes before, we propose to use a PPS with space division multiplexing expansion (SDME PPS for short) to achieve high performance to support the multicast traffic while only requiring k redundant multicast modules. Our proposal is highly scalable and simple to implement, since it not only eliminates the unpractical speedup requirement but also keeps scalability advantage in the same way as traditional PPS schemes.