From: Michelle Liu (liujing@u.washington.edu)
Date: Tue Oct 19 2004 - 21:38:14 PDT
Review of "Congestion Control for High Bandwidth-Delay Product Networks"
Jing Liu
This paper proposes a protocol named as eXplicit Control Protocol (XCP), which is better than TCP in managing congestion control in networks. XCP outperforms TCP in both conventional and high bandwidth-delay environments. Furthermore, XCP achieves fair bandwidth allocation, high utilization, small standing queue size and near-zero packet drops.
TCP works well with current networks. However, as the product of link bandwidth and delay increases, TCP congestion control mechanisms have some drawbacks, such as oscillation, instability, low utilization of bandwidth and unfairness etc. In this paper, the proposed XCP fixes some of the problems of TCP. The strong points of XCP lie in the following. First, XCP uses explicit congestion feedback while TCP uses packet loss as a signal of congestion. In this way, XCP is both more responsive and less oscillatory. The on time feedback enables the packet drop to be close to zero. Second, the routers of XCP have both an efficiency controller (EC) and a fairness controller (FC). The EC aims at maximizing the utilization of link bandwidth and FC aims at fair bandwidth allocation. In addition, the parameters of EC and FC are independent of the number of flows. This provides much flexibility for protocol design. In TCP, additive-increase multiplicative-decrease (AIMD) is used to adjust the window size. While in XCP, EC uses multiplicative-increase multiplicative-decrease (MIMD) and FC uses additive-increase multiplicative-decrease (AIMD). The former allows XCP to quickly acquire the positive spare bandwidth even over high capacity links and the latter converges to fairness.
The major contribution of XCP is that it achieves some QoS and also helps dealing with the security issue by explicit feedback.
Some of weaknesses could be. First, there is much overhead for the routers, since the routers must calculate whether the feedback is positive or negative and EC and FC are both implemented in the routers. Second, the author mentions that the routers have to monitor the input traffic rate to each of their output queues in order to decide whether there is congestion happened or not. However, there is nothing mentioned in the paper that how the input traffic rate is measured. Third, in order to achieve low packet dropping rate, XCP prevent, as much as possible, the queue from building up to the point at which a packet has to be dropped. This might lower the queue's utilization or a waste of memory in the routers.
Considering that the link bandwidth increases dramatically in the Internet, to provide some QoS and taking care of security issues is necessary in the future. This paper is working towards this direction.
This archive was generated by hypermail 2.1.6 : Tue Oct 19 2004 - 21:38:15 PDT