MIXD-TCP: Completely decoupled end to end congestion control algorithm

Abstract

Transmission Control Protocol (TCP) is most widely used protocol in the Internet. In order to overcome inefficiency of TCP's Additive Increase and Multiplicative Decrease (AIMD) policy, several end-to-end algorithms have been proposed. However, these algorithms need considerably large time for converging to efficiency and fairness. In this paper, we propose Multiplicative Increase and Exponential Decrease transmission control protocol (MIXD-TCP). MIXD-TCP is a distributed and end-to-end protocol, i.e. each user updates its own transmission rate based on one bit feedback from the receiver indicating packet drops. MIXD-TCP decouples efficiency and fairness while at the same time converges to ‘efficient and fair resource allocation’ in the network. We prove that this optimal allocation is globally asymptotically stable under MIXD-TCP. Using ns-2 simulations, we demonstrate that MIXD-TCP performs better than existing end-to-end algorithms.