From Measurements to Modeling The Trade-off between Energy Efficiency and System Performance

In this thesis, the work is based on experimental and modeling methodologies. I applied this starting from measurements of Asymmetric Digital Subscriber Line (ADSL) technology, which is the preferred high-speed access to the Internet. The work focuses on the problem of determining the maximum stable bandwidth that an Internet Service Provider (ISP) can offer. This task was accomplished by monitoring and analyzing a large set of ADSL end-users. Huge amount of collected data allowed us to get a detailed statistical analysis of the behavior of ADSL lines, such as the distribution, variable dependencies and correlation of effective bitrate with the physical measurements exposed by network devices. Analysis of collected data suggest that there is no clear trend to predict a downstream bandwidth of ADSL line based on line conditions, and thus requires an intelligent way of analyzing the ADSL lines. Thus, a Neural Network (NN) was employed, which is an intelligent machine learning tool to (i) learn the behavior of ADSL lines, (ii) extract useful information from huge set of measurements, (iii) automatically suggest maximum stable bandwidth. The results indicate that NN performs well in predicting end-users available bandwidth. However, NN is required to be properly trained, as well as needs careful selection of design parameters. Later, the focus of the work was centered over the energy efficiency of telecommunication systems using mathematical modeling approach. Motivation of first work was to know how much energy efficient is Voice over Internet Protocol (VoIP) architecture as compared to traditional Public-Switched Telephone Network (PSTN) architecture used for voice communication. To answer this, these two architectures already implemented at our campus were extensively examined by means of measuring real power consumption. To generalize or estimate power consumption for any given number of users, a mathematical model of power consumption for both the architectures is built. The results indicate that VoIP architecture consumes a lot of power, yet VoIP system have the flexibility to be made energy efficient by adding some sort of energy-wise schemes. The last part of the work investigates the interaction between the Transmission Control Protocol (TCP) and rate adaptation mechanism. Where rate adaptation is an approach to save energy by adapting the transmission rates or capacity according to traffic dynamics. To have deep insight of the interaction, a mathematical model based on fluid based framework is built to depict the behavior of TCP and rate adaptation scheme. The model is then tested for its accuracy and stability by conducting simulation experiments and steady state analysis. Later, the model is used to study the impact of tuning the parameters on system performance. The results suggest that the implementation of rate adaptation scheme can be effective but a careful parameter setting is needed to avoid undesired disruptive interaction among controllers at different levels, that impair QoS.