Multicast scheduling in feedback-based two-stage switch

Abstract

Scalability is of paramount importance in high-speed switch design. Two limiting factors are the complexity of switch fabric and the need for a sophisticated central scheduler. In this paper, we focus on designing a scalable multicast switch. Given the fact that the majority traffic on the Internet is unicast, a costeffective solution is to adopt a unicast switch fabric for handling both unicast and multicast traffic. Unlike existing approaches, we choose to base our multicast switch design on the load-balanced two-stage switch architecture because it does not require a central scheduler, and its unicast switch fabric only needs to realize N switch configurations. Specifically, we adopt the feedback-based two-stage switch architecture [10], because it elegantly solves the notorious packet mis-sequencing problem, and yet renders an excellent throughput-delay performance. By slightly modifying the operation of the original feedback-based two-stage switch, a simple distributed multicast scheduling algorithm is proposed. Simulation results show that with packet duplication at both input ports and middle-stage ports, the proposed multicast scheduling algorithm significantly cuts down the average packet delay and delay variation among different copies of the same multicast packet.