Queuing analysis and optimization techniques for energy efficiency in packet networks

摘要

Energy efficiency in all aspects of human life has become a major concern, due to its significant environmental impact as well as its economic importance. Information andudCommunication Technology (ICT) plays a dual role in this; not only does it constituteuda major consumer itself (estimated 2-10% of the global consumption), but is also expected to enable global energy efficiency through new technologies tightly dependent onudnetworks (smart grid, smart homes, cloud computing etc.). To this purpose, this workudstudies the problem of energy efficiency in wired networks. As this subject has recentlyudbecome very active in the research community, there is parallel research towards severaludresearch directions. In this work, the problem is being examined from its foundationsudand a solid analytical approach is presented.udSpecifically, a network model based on G-network queuing theory is built, whichudcan incorporate all the important parameters of power consumption together with traditional performance metrics and routing control capability. This generalized modeludcan be applied for any network case to build optimization algorithms and estimate theudperformance of different policies and network designs. Composite optimization goalsudfunctions are proposed, comprising both power consumption and performance metrics.udA gradient descent optimization algorithm that can run in O(N3) time complexity isudbuilt thereof. Using power consumption characteristics of current and future equipment,udseveral case studies are presented and the optimization results are evaluated. Moreover,uda faster gradient-descent based heuristic and a decentralized algorithm are proposed.udApart from the routing control analysis, the case of a harsher energy saving solution,udnamely turning o the networking equipment, is also experimentally explored. Applyinguda tradeoff study on a laboratory testbed, implementation challenges are identified andudconclusions significant for future work are drawn. Finally, a novel admission controludmechanism is proposed and experimentally evaluated, which can monitor and manageudthe power consumption and performance of a network.