Proc. Eighth Asia-Pacific Computer Systems Architecture Conference, Aizu-Wakamatsu City, Japan, September 2003, pp 380-394 doi:

Philip Machanick and Brynn Andrew

Virtual multicasting (VMC) combines some of the benefits of caching (transparency, dynamic adaptation to workload) and multicasting (reducing duplicated traffic). Virtual multicasting is intended to save bandwidth in cases of high load, resulting from unpredictable but high demands for similar traffic. However, even in cases where relatively low fractions of traffic are similar (hence o ering few opportunities for VMC), introducing VMC can have a disproportionate e ect on latency reduction because of the generally bene cial e ect of reduction in traffic, including reduced contention. This paper presents results of a study of latency reduction across a range of workloads, illustrating the potential for VMC even in situations where the extent of overlapped traffic is light.

© Springer-Verlag 2003; published as a volume of LNCS

(PDF and HTML at Springer web site: subscription required; local PDF 437K)