Implementation and Analysis of the TCP "Adaptive-Selection" Concept in ns-2 and Linux

The increasing level of heterogeneity of present and future networks poses new challenges to TCP. Wired and wireless connections (including satellite) with very different characteristics, concerning bandwidth, RTT and loss rate, will have to co-exist and to compete, possibly in a fair way, in the same network. This heterogeneity makes difficult the choice between the many TCP enhancement proposals, as they basically are focused on single impairments, resulting "optimized" for different network environments. To overcome this problem the authors have recently presented the TCP-adaptive selection (TCP-AS) concept, which challenges the very idea that one choice should actually be made. This paper aims at presenting the implementations of the TCP-AS in both Linux and ns-2. They leverage on the modular structure of the TCP Linux congestion control, which can be replicated in ns-2 by making use of the Caltech "NS-2 TCP-Linux" patch. As described in the paper, the implementation key is the insertion of an intermediate agent, called "tcpas", between the TCP monolithic core and the TCP congestion control modules. In the second part of the paper, results of the ns-2 implementation have been cross-checked with Linux results. The analysis of data confirms both the feasibility and the performance advantages of the TCP-AS approach.