Branch-point algorithms for multicasting ATM ABR protocols

The ATM ABR traffic management entails that the source controlled to the minimum supported by all the leaves of the multicast. Since the amount of feedback (i.e., the number of backward resource management or BRM cells) received by the source is proportional to the number of leaves (destinations) in the multicast tree, it is necessary for the branch points to properly consolidate feedback to avoid feedback impulsion and feedback noise problems. We first examine the feedback impulsion and scalability issues in two major groups of branch-point algorithms appeared in the literature: one requires switch points to turn around FRM cells, and the other requires switch points to consolidate and pass BRM cells back to the source. We formally analyze these two groups, and show the exact differences of these two groups in BRM-cell bandwidth overhead, BRM cell inter-arrival time at the source, and the end-to-end delay experienced by BRM cells. The results indicate that the two groups result in similar feedback impulsion, and have the same degree of scalability in terms of BRM cell feedback delay. We study and compare three branch-point algorithms, proposed by the authors and by (1) Tzeng and Siu (see ACM SIGCOMM Computer Communication Review, p.81-106, 1995 and IEEE JSAC, vol.15, no.33, p.545-56, 1997), and (2) Fahmy, Jain, et. al. (see Proc. of IEEE INFOCOM, San Francisco, p.1004-13, 1998). Through simulation, we evaluate the effects of network configurations, underlying multicast algorithms, and multicast traffic distribution on their performance. We found that the underlying unicast algorithm has a major impact on traffic flow when unicast traffic is dominant, regardless of the branch-point algorithm. When multicast traffic is dominant, the branch-point algorithm plays a more significant role. We believe that this is one of the first papers to investigate the effect of the unicast algorithm on multicast flow control.