"... In PAGE 22: ...Table 1: Comparison of algorithms for adjusting bandwidth in response to congestion In this section, we categorize approaches towards bandwidth adaptation in reponse to congestion, as summarized in Table1 . The first row of Table 1 shows approaches that rely on reservation, and the second row shows approaches that do not.... ..."

"... In PAGE 19: ... 8.3 Results Table1 shows estimated link and surplus bandwidths for measurements where the network has one con- gestible link. These measurements show the accuracy of our analysis method.... ..."

"... In PAGE 16: ... In order to facilitate the presentation of results for the different possible variations of parameters, simulations were numbered according to their combination of parameters. This numbering is summarized in Table1 . Based on this numbering, the simulation parameters for a particular simulation can be ascertained from its ID.... ..."

"... In PAGE 8: ... from 0:05 Mbps to 1 Mbps. Table2 lists the loss rates at link (r0; r1) for various bandwidths settings. No loss occurs at other links.... In PAGE 8: ...y Proposition 2.1. We further estimate an upper bound on the maximum queuing delay of the strongly dominant link, which is link (r0; r1). Table2 lists the actual maxi- mum queuing delay of link (r0; r1), and the estimates from our approach and the loss pair approach. Estimates from both approaches are close to the actual value in all the set- tings: the maximum error from our approach and the loss pair approach is 2 ms and 5 ms respectively.... In PAGE 8: ... Estimates from both approaches are close to the actual value in all the set- tings: the maximum error from our approach and the loss pair approach is 2 ms and 5 ms respectively. We only describe one setting in Table2 in detail. In this setting, the bandwidth of link (r0; r1) is 1 Mbps.... ..."

"... In PAGE 20: ... We would like to point out that the accuracy of the subsequent link estimations is still a matter of further research, but we present these measurements here for two reasons; to show that we have no problem at estimating the first link in these scenarios, and also that the estimation of subsequent links can be quite accurate as well. Table2 shows estimations for two congestible links. The two links in this scenario are one 34 Mbps link with 3 Mbps surplus and one 10 Mbps link with 7 Mbps surplus, i.... ..."

"... In PAGE 7: ... For each topology, we compare the congestion counts in Experiment B through D with the congestion count in Experiment A, and get the ratios (Cx=CA) that represent the percentages of congestion counts still left after applying a certain heuristics or all of them. Table2 shows the average result after running simulation on 30 random multicast trees (10 networks each for maximum node degrees of 3,4 and 5).... ..."

"... In PAGE 9: ...re con rmed by Proposition 2.2. We estimate an upper bound on the maximum queuing delay at the weakly dominant congested link, which is link (r2; r3). Table3 lists the actual maximum queuing delay of link (r2; r3), and the estimates from our approach and the loss pair approach. The estimates from our approach are much closer to the actual value (with a maximum error of 9 ms) than those from the loss pair approach (with a maxi- mum error of 100 ms).... In PAGE 9: ... The reason for the deviation from the loss pair approach might be that this approach is sensitive to the queuing delays at links other than the weakly dominant congested link and/or requires longer probing durations. We next describe one setting in Table3 in detail. In this setting, the bandwidths of link (r0; r1) and (r2; r3) are 0:7 Mbps and 0:2 Mbps respectively.... ..."