Insights into TCP/IP congestion control using the smith principle

Abstract

In this paper we propose classical control theory to look at the end-to-end congestion control algorithm used by the Transmission Control Protocol/Internet Protocol (TCP/IP). By using this approach we show that classical control modeling using Laplace transforms and Smith's principle give a simple and rigorous model of the role played by congestion window and advertised window variables in today's in TCP/IP. In particular, we clearly show that TCP performs flow and congestion control using a Smith's predictor. The proposed analysis gives clear guidelines to improve TCP in a way that is backward compatible and incrementally déployable. In particular, the model shows how it is possible to include in the control scheme, which has been obtained using a Smith predictor, any new type of explicit feedback from network nodes, any new algorithm executed at the source, as well as any new implicit feedback. Following this approach a modification to TCP called faster recovery is briefly sketched.