Research On Optimization Algorithms For Reducing Energy Consumption Of Data Centers And Improving Quality Of Service Of The Network

The energy consumption of networking elements in modern cloud computing datacenters accounts for a large part of its total power consumption. With the developmentof virtual machine technique and programming switch technology, more and morecontrol strategies and optimization methods have been proposed to reduce the energyconsumption of networking elements and improve the quality of services of the datacenter network. In this thesis, I firstly introduce and analyze the virtual machinetechnology and the characteristics of the communication traffic flows in cloudcomputing data centers. And then, with regard to the two main services in moderncloud computing data centers, which are common internet business handled byMapReduce and kinds of scientific workflows, I propose “advanced geneticalgorithm” and “job-aware VM placement and route scheduling” algorithm (JAVPRS)respectively.Considering the huge amount of virtual machines and physical machines inmodern data centers, I propose a novel gene encoding method for the geneticalgorithm which is for optimizing the placement of virtual machines carryingMapReuce jobs. The gene encoding method is a two-dimensional array composed byidentifier of virtual machines and physical machines. The crossover operation of thegenetic algorithm is base on this kind of encoding method, and the cutting point is themiddle of the two independent dimensions. This kind of crossover operation not onlypersist the coupling property of the virtual machines, but also guarantee the solutionspace traversal.In the JAVPRS algorithm for scientific computing jobs, I migrate the jobs whosecommunication time is larger than the time threshold (TH) to “neighbour” physicalmachines according to the time feature of communication data flows, and consolidatethe communication data flows into part of the communication links in order to reducethe amount of working switches and balance the communication load in powered onswitches and links.The main purpose of these two optimization algorithm is to reduce the energyconsumption of the networking elements, improve the utilization of the network bandwidth, and promote the QoS of the data center network. At last, this thesis make asimulation in NS2for a Fat-tree topology data center. The results suggest that theenergy consumption of networking elements and the loss rate of communicating datapackets can be reduced significantly, and the network throughput can be improvedgreatly. The data delay can be reduced2.6%at least for MapReduce jobs whenapplying the genetic algorithm to optimize the placement of the virtual machines. Forthe scientific computing jobs applied the JAVPRS algorithm to optimize theplacement of the virtual machines, the energy consumption of the networkingelements can be reduced3.1%at least, the data delay can be reduced8.1%at most,and the network throughput can be improved11%at most. The results demonstratethat the two optimization algorithms can reduce the energy consumption ofnetworking elements and improve the quality of services of the data center networkthrough optimizing the placement of VMs and dynamically assigning the route pathsfor traffic flows.