Improving congestion control performance through loss differentiation

Linear increase/multiplicative decrease (LIMD) has typically been the congestion control paradigm of choice in the Internet. However, a major drawback of LIMD is that it reacts identically (and aggressively) to all packet losses, irrespective of the cause of loss. In this paper, we try to augment the basic LIMD congestion control with additional mechanisms to predict the cause of packet losses and react accordingly. To this end, we present the LIMD/H algorithm, which has the following features: (a) LIMD/H uses the “history” of packet losses and the evolution of transmission rate for a connection in order to distinguish between congestion-induced and non-congestion-induced packet losses; (b) LIMD/H reacts gently to non-congestion-induced losses and aggressively to congestion-induced losses, thereby achieving high efficiency, fairness, as well as quick reaction to the onset of congestion. We present an initial performance evaluation of LIMD/H using simulations and analysis