I. Introduction
Much research has been conducted to achieve stable, efficient and fair operation of packet-switching networks. TCP congestion control [16], a set of end-to-end mechanisms, has been widely acknowledged for its critical role in maintaining stability of the Internet. Among them, TCP Reno [2] infers network congestion by detecting packet loss that is presumably caused only by congestion; alternatively, TCP Vegas [10] measures backlog, the number of buffered packets inside congested routers along the path, to detect network congestion and demonstrates better performance than Reno. Unfortunately, Vegas has technical problems inherent to its backlog estimator that prevent it from functioning properly. There has been a substantial amount of work on this issue, such as [1], [8], [9], [12], [15], [24], [25], and the references therein, which we review in Section II. But none of them provides a solution to measuring backlog unbiasedly in case of round trip propagation delay estimation error or reverse path congestion.