Computational Energy Cost of TCP in MANETs

In this paper, we present the results from a detailed energy measurement study of different TCP variants when used in Mobile Ad hoc Network environments. More precisely, we focus on the node-level cost of the TCP protocol; also know as the computational energy cost. In fact, the computational energy consumption is the most important part of TCP energy consumption. This is already proven in previous work and our results confirm this fact. Sometimes, the computational energy cost is three times that of the communication energy cost. The studied TCP variants, in this work, are TCP New-Reno, Vegas, SACK, and Westwood. In our analysis, we draw a breakdown of the energy cost of the main congestion control algorithm (i.e. slow start, fast retransmit/fast recovery, and congestion avoidance) used by these TCP variants. The computational energy cost is studied using a hybrid approach, simulation/emulation, using the SEDLANE emulation tool. This study takes into consideration different data packet loss models (congestion, link loss, wireless signal loss, interference) within such environments when different ad-hoc routing protocols (reactive and proactive) are used. The performed study gives a set of results that are of high interest for future improvements of TCP in MANETs. Among the obtained results, we show that the computational energy cost of TCP varies according to the type of data packet loss model it comes through: network congestion, interference, link loss, or signal loss. The results demonstrate that the link loss scenario is the most severe situation for TCP connections to face. In addition to that, we show that the Fast Retransmit/Fast Recovery phase has much less energy cost than both Slow Start and Congestion Avoidance phases, due to the fact that it sends more TCP data bytes in a shorter period of time. Finally, the computational energy cost is quantified and compared to the TCP end to end performance for each TCP variant showing the link between both.