Researches On High-efficiency Resources Allocation Algorithms In Cloud Computing Data Center

With the popularity of the Internet and intelligent terminals,the global informatization level is increasing,and the demand for the storage and processing of massive users' data has been rising continuously,and then Cloud computing data centers have become an important infrastructure platform in the information Era,Data centers integrates huge amounts of computing,storage and communication resources,and provides important supports for the storage,processing and application of massive data,keeping prompting the sustainable development of the Internet.According to Gartner's prediction,the number of global data centers will reach 422,000 by 2020.However,the high energy consumption problem has severely restrained the development of cloud data centers.Since the total energy consumption of nationwide data centers will exceed 200 billion k Wh between 2020 and 2021,many data center construction plans cannot be approved at present.In data centers,the low resource utilization of IT(Information Technology)equipment is one of the key factors resulting in the high energy consumption problem.The low utilization of IT equipment resources results from the contradiction between the time-varying characteristics of computing,storage and communication demands and the lack of flexibility in resource scheduling algorithms.At present,the embedding of computing and communication resources in cloud data centers are facing the following problems.1)The joint placement of multiple virtual machines is not flexible enough.A large number of computing services change dynamically over time,but the existing virtual machine placement methods seldomly took into account the time-varying characteristics of computing services,which makes the differentiated time-varying characteristics of computing services absent in virtual machine joint placement.Therefore,the traditional virtual machine placement methods usually were based on services' maximal resource demand and resulted in low computing resource utilization.2)The fix topology of data center networks is hard to be scheduled flexibly,and highly redundant communication resource is deployed to handle the links' peak traffic.By introducing wireless links into the wired network structure,the network topology can be dynamically constructed to adapt to the distribution of communication traffic.However,how to jointly schedule computing resources and wired/wireless communication resources to improve the network resource utilization remains to be solved.Focused on those above two problems,the main contributions of this thesis are listed as follows.1.A multiple virtual machine joint placement method based on the time-varying characteristics of computing services is proposed,which would effectively improve the computing resource utilization.First,the correlation coefficient of computing services is defined to represent the correlation of resource demands from different computing services over time.The more complementary two computing services are,the smaller correlation coefficient they have.Second,a smoother integrated computing service can be obtained after multiple computing services with high complementarity are jointly mapped to the same physical machine.Then the difference between the peak and valley of the integrated computing services are reduced,the computing resource utilizations of the physical machines can be effectively improved,and the number of active physical machines can be also reduced,which can improve the cloud data center's energy efficiency.Next,aimed to solve the NP-hard joint placement problem,a computing load based sequential correlation matching algorithm is designed,which gives priority to these computing services with higher revenues.This method can both promote the virtual machine placement revenue and guarantee the virtual request acceptance ratio performance.2.Aimed at the wired and wireless hybrid network structure,this thesis proposes a computingcommunication joint virtual network mapping scheme based on minimized resource occupation,which can effectively improve the resource utilization ratio of embedding virtual networks to the substrate networks.A two-stage mapping method is designed to solve the integer programming problem of joint mapping.In the first stage,the computing virtual nodes are sorted decreasingly according to their communication demands,and then the virtual nodes will be mapped integrally as a whole or mapped into different but close deployed physical machines based on those nodes' communication demands,with the cloud data center's topology and wireless link capacity as constraints.In the second stage,aiming at minimizing the occupancy rate of communication resources,we take full advantage of the flexible deployment feature of wireless links and construct the wired and wireless hybrid paths to guarantee the communication demand between virtual nodes,while minimizing the occupancy of communication resources.Simulation results have proved that the propose scheme can achieve higher resource utilization when mapping virtual networks to physical networks.