Analytical and experimental evaluation of TCP with an Additive Increase Smooth Decrease (AISD) strategy

Abstract

This paper presents an analytical model to study the throughput performance of TCP with an Additive Increase Smooth Decrease (AISD) strategy. The new AISD-based TCP protocol produces smooth transfer rates and performs the flow and congestion control functions of TCP. As a result, it is suitable for streaming applications and solves the unfairness problem that occurs when TCP and UDP share the same bottleneck link. The AISD strategy is very simple to implement, only modifies the multiplicative part of the Additive Increase Multiplicative Decrease strategy of TCP, and can be implemented in any TCP version. The smooth part of the strategy is implemented using a low pass filter that considers history in the calculation of the congestion window. The modeling of this new strategy raises new challenges compared to the classical TCP modeling in two ways: first, it needs to be adapted to a more complex dynamism of the congestion window, and second, the model needs to incorporate a scheduler that periodically updates the value of the congestion window. The model is progressively built to finally characterize the steady-state send rate and throughput of a flow as a function of the loss probability, the round-trip time (RTT), the time-out interval, and the scheduler interval. The performance of this AISD-based TCP is compared analytically and experimentally with TCP Reno, and its superior performance is demonstrated.