IP Queueing Delay Applied to TCP Congestion Control

Abstract

TCPs congestion control was traditionally based on packet losses. This approach leads to an inefficient bandwidth use, because losses must be produced to infer network load. In last years, another approaches have been considered to infer that load, mainly the estimation of sender-receiver delay, known as ” one way delay” (OWD). The estimation of this delay enables the sender anticipate losses and lower its sending rate before they occur. A known problem is to accurately estimate OWD, because it requires synchronization between sender and receiver. Various alternatives exist to achieve synchronization (NTP, GPS), but these are not applicable to TCP’s flows due to the significant number of flows and because they mainly reside on users’ equipment. Another approach that has been recently incorporated in T C P ’s flow control mechanisms, consists in infer OWD from the RTT value by means of some heuristic. However, this approach has lead to develop complex heuristics that produce inaccurate results, due mainly to the assymetry in the paths and the variability of network load. In this paper a simple mechanism to estimate OWD is proposed. It is based in the cooperation of the network layer, and consists in obtain OWD as the sum of queueing delays experienced by packets in each router along the path from sender toward receiver.