Date of Award

Summer 2006

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical/Computer Engineering

Committee Director

Min Song

Committee Member

James Leathrum, Jr.

Committee Member

K. Vijayan Asari

Committee Member

Hussein Abdel-Wahab

Abstract

Due to its mild requirement on the bandwidth of switching fabric and internal memory, the input-queued architecture is a practical solution for today's very high-speed switches. One of the notoriously difficult problems in the design of input-queued switches with very high link rates is the high performance queueing and scheduling of multicast traffic. This dissertation focuses on proposing novel solutions for this problem. The design challenge stems from the nature of multicast traffic, i.e., a multicast packet typically has multiple destinations. On the one hand, this nature makes queueing and scheduling of multicast traffic much more difficult than that of unicast traffic. For example, virtual output queueing is widely used to completely avoid the head-of-line blocking and achieve 100% throughput for unicast traffic. Nevertheless, the exhaustive, multicast virtual output queueing is impractical and results in out-of-order delivery. On the other hand, in spite of extensive studies in the context of either pure unicast traffic or pure multicast traffic, the results from a study in one context are not applicable to the other context due to the difference between the natures of unicast and multicast traffic. The design of integrated scheduling for both types of traffic remains an open issue.

The main contribution of this dissertation is twofold: firstly, the performance of an interesting approach to efficiently mitigate head-of-line blocking for multicast traffic is theoretically analyzed; secondly, two novel algorithms are proposed to efficiently integrate unicast and multicast scheduling within one switching fabric.

The research work presented in this dissertation concludes that (1) a small number of queues are sufficient to maximize the saturation throughput and delay performances of a large multicast switch with multiple first-in-first-out queues per input port; (2) the theoretical analysis results are indeed valid for practical large-sized switches; (3) for a large M × N multicast switch, the final achievable saturation throughput decreases as the ratio of M/N decreases; (4) and the two proposed integration algorithms exhibit promising performances in terms of saturation throughput, delay, and packet loss ratio under both uniform Bernoulli and uniform bursty traffic.

DOI

10.25777/3hyz-5p33

ISBN

9780542896972

Share

COinS