Research On Power-aware Routing In Data Center Networks

In order to meet the high network bandwidth and reliability requirementfrom cloud computing applications, many advanced data center networkarchitectures have been proposed recently. They use abundant networkresources to interconnect servers and provide high-performance andfault-tolerant routing and switching services. However, these “richly-connected”network architectures suffer from skyrocketing energy consumption andinefficient energy usage, which have aggravated the operational cost and energywaste of data center networks. In this dissertation, we analyze the networkenergy efficiency of typical data center network architectures. We study how toreduce the energy consumption and improve the energy usage efficiency of datacenter networks from the routing and flow scheduling perspective. The majorwork and contributions of this dissertation are as follows:(1) A survey of network power-saving technologies. We present theclassification principle and research method of network power-savingtechnologies. We analyze and evaluate the related problem models,implementation details and technology limitations of typical power-savingschemes.(2) A study of power-aware routing model in data center networks. Weanalyze the power consumption characteristic of modular network devices, andpresent the power consumption computation models, which can be used toevaluate the effectiveness of power-saving schemes. We propose thepower-aware routing model in data center networks and use it to guide thepower-aware routing and flow scheduling design.(3) Energy efficiency analysis of data center networks. We propose acomparison method of network energy efficiency among different data centerarchitectures. We analyze the network energy consumption characteristic, theeffectiveness of power-saving schemes and the tradeoff between networkperformance and energy consumption in typical data center architectures. Weshow the comparison results of network energy efficiency in these architectures, which can greatly help operators when designing and upgrading data centernetworks.(4) Throughput-guaranteed power-aware routing. We formally establish theproblem model, and prove that it is NP-Hard. We propose an effectivepower-aware routing algorithm, which calculates power-saving transmissionpaths for network flows by pruning the network topology, while satisfying therestrictions of network throughput and reliability. The simulation results showthat the scheme can effectively reduce the energy consumption of data centernetworks, especially under the low network load.(5) A joint power-saving scheme of routing and flow scheduling. We studythe impact of network flow scheduling and routing strategies on the energyconsumption of data center networks. We propose a joint scheme of flowpreemption and power-aware routing. It enables the traffic flows to exclusivelyoccupy the bandwidth of links on their transmission paths, and thus increasesthe utilization ratio of links and network energy usage efficiency. The schemecan effectively save network energy and improve average completion time ofnetwork flows.