The ABR congestion control by the least squares estimation of spatial distribution of sources

Abstract

In this paper, control of ATM ABR traffic is considered for a simple system with a single bottleneck switch and several sources at different distances from the switch. User population is partitioned into delay zones according to the round trip delays between the switch and the zones. The switch employs an ER (explicit rate) control. The contribution of input traffic from a source is upperbounded by the ER, which was specified by the switch round trip delay earlier. The aggregate input rate into the switch then becomes a weighted sum of past ER's, where the weights are determined by the number of sources and their levels of activities in respective zones. Given the history of aggregate input rates and ER's, the switch is able to estimate the traffic intensities from individual zones. The results from several scenarios demonstrate that the least squares estimation of the weight distribution across zones can provide the convergence in the maximum round trip delay.