Research On Optimizing Virtual Machine Placement In Cloud Computing Data Centers

Cloud computing is a new model of computing and service.It integrates a large number of resources,such as servers and storage devices,into a huge pool of resources,and provides the on-demand,self-help and extensible services to tenants(or cloud users)through cloud platform and Internet.In recent years,cloud computing has developed rapidly,and it becomes more acceptable.More and more applications and services are migrated into cloud data centers.For the cloud providers,the cloud resources in their cloud data centers are limited and unevenly distributed,and an appropriate placement of cloud users' servers and data is crucial to improve user experience.In the modern cloud data centers,virtual machine(VM)placement technology is the primary issue faced by cloud providers to efficiently schedule cloud resources.In addition,for the cloud users(or tenants),their budgets of deploying application on cloud systems are limited,which leads to the limited number and location of their servers.Thus,the network performance of the applications and services are decided by the deployment scheme of their servers.Based on the existing VM placement technologies in cloud computing,this disser-tation studies the VM(server)placement scheme for cloud providers and cloud users to improve the revenue of cloud systems and reduce the request response latency,re-spectively.For the cloud providers,the statistic model of the applications' requests and the scalability of cloud data centers are taken into account,respectively.For the cloud users,the request routing strategy of their large-scale applications is considered.The main contributions of this dissertation are as follows:1.Based on the statistic model of applications' requests and the service model of VMs with data nodes,the VM placement technology in clouds is studied to re-duce the response latency of requests and improve the user experience of cloud systems.The tenant's requests are modeled as independent Poisson streams,and the VMs with data nodes are modeled as M/M/l queueing systems.Then,this dissertation propose two optimization objectives for VM placement:Minimizing Maximum Latency(MML)and Minimizing Total Latency(MTL).For the objec-tive of MML,an(1+?)-approximation algorithm is proposed.For the objective of MTL,this dissertation propose three algorithms.In addition,in the condi-tion of lacking of resources,this dissertation propose two extension algorithms to optimize the objective MML and MTL respectively,while ensuring the num-ber of deployment applications is maximized.The simulation results show that the proposed model and algorithms can efficiently reduce the response latency of requests.2.For expanding the clouds,a cloud system model for cloud providers is proposed to dynamically expand the geographically distributed cloud data centers.In this model,the cloud providers can expand its cloud data centers by renting physical resources from other Resource Owners(ROs).This dissertation define two ser-vice modes for the cloud providers.Given the set of ROs with their resources,and the set of cloud users with their VM requirements,the expanding geo-distributed cloud problem is to determine the rented resources and the VM placement scheme such that the revenue of the cloud system is maximized.Two polynomial-time heuristic algorithms are proposed to solve the problems in two modes respec-tively.This dissertation consider a more general leasing strategy,which is called part-leasing.Under this strategy,this dissertation also propose two extended al-gorithms to solve the problem.Our simulation results show that the system model and algorithms can effectively improve the user satisfaction and the total revenue and reduce the average latency of cloud users' requests.3.For the deployment of large-scale application in geographically distributed cloud data centers,this dissertation consider the application's request routing strategy.This dissertation propose a congestion game-based interpretation of the request routing strategy and define the request response latency as queuing time in VM.Given a large-scale application with its estimated parameters,and a budget of the application provider,the VM placement problem is to find a VM(server)placement in a set of cloud datacenters such that the total latency is minimized.This dissertation formulate this problem as a mixed integer bi-level optimization program.By utilizing the structure of the congestion game's Nash equilibrium,an equivalent single-level optimization program is successfully obtained and a 2-approximation algorithm is proposed.This dissertation also consider the VM placement problem for multiple applications,and propose a 2-approximation al-gorithm.The simulation results show that the proposed model and algorithms significantly outperform the previous works.