On the impact of the dropping function on the packet queueing performance

Abstract

To eliminate the bufferbloat phenomenon from the Internet, it is recommended to apply the active queue management in network nodes — for instance, by dropping packets with a probability that is a function of the queue size (dropping function). The positive effects of this mechanism on the queue sizes, queueing delays, TCP flow desynchronization and inter-flow fairness, have been studied in many papers. They did not deal, however, with the statistical structure of packet losses caused by the application of the dropping function. This structure is especially important when real-time time multimedia transmissions are considered — such transmissions are very sensitive to consecutive packet losses. In this paper, we study the impact of the dropping function on the burst ratio, a formally-defined measure of the tendency of packet losses to occur in long series. We also check the tradeoff between the burst ratio and the overall packet loss ratio, using the cost function proposed for voice transmissions. Finally, taking this tradeoff into account, we draw conclusions about best shapes of the dropping function.