Research On Virtual Machine Placement In Datacenter

With the increasing development of computer network,cloud computer is one of the most interesting technologies.Cloud computing has attracted many industry and academia people's concern,due to flexible,convenience and on-demand nature of resource provisioning.As the cloud computing infrastructure,the datacenter also let many IT providers begin to pay close attention to and build their own datacenter,such as Google,Microsoft,Amazon and Oracle.Existing studies show that a reasonable virtual machine(VM)placement method can ef-fectively improve the resource utilization,decrease the total task completion time of datacenter,improve the datacenter's the total throughput of the execution tasks,etc.In this dissertation,VM placement method is divided into dynamic placement and static placement.Dynamic placement,namely migration,is that VMs change the location among the physical machines(PMs)as the datacenter system may require when VMs execute tasks.Static placement,namely initialization placement,is that each VM is placed in PM or in the datacenter system by requirement before VMs execute tasks.The VMs do not change the location before tasks are completed.Based on a comprehensive analysis of datacenter's approach of VM placement and related works,this dissertation focuses on VM optimizing placement in datacenter,and gains several achievements on some sub-topics.The major contributions of this dissertation are as follows:1.Improve the PM resource utilization and minimize the cost of VM migration based on VM dynamic placement.With the increasing development of the datacenter as well as bring a lot of problems,two main of which are that how to improve datacenter resources utilization and how to reduce the datacenter VM migration costs.While recent studies have primarily focused on maximizing PM resource utilization or minimizing the cost of VM migration separately,the single target optimization method cannot have satisfied the datacenter which needs for high-use PM resource and high-use bandwidth resource.However,there has been little attention on jointly taking these two objectives into account.Therefore,in this dissertation,we present an optimization model for the datacenter of high-use PM and high-use bandwidth.The model takes into account minimum VM migration cost and maximum PM resource utilization with multi-resources such as memory,bandwidth,CPU,and disk space.The optimization of our proposed model is non-deterministic polynomial-time hard.Therefore,we present an efficient approximate algorithm based on bin packing algorithm,called MinCost,to resolve our proposed model and obtain a near-optimal solution.Finally,the simulation results in this dissertation show that our proposed model and algorithm can get the maximizing PM resource utilization and minimizing VM migration cost.2.Decrease the total task completion time of datacenter based on VM static placement.The datacenter is divided into data-intensive and computing-intensive.In the computing-intensive ar-chitecture,the time of access latency has a great influence on completion time of task.The data transfer time(DTT)is one of dominating factors of task completion time in the data-intensive architecture.Existing studies show that a reasonable VM static placement method can effec-tively resolve the above problems.A good optimal VM static placement result can obtain the short time of access latency in computing-intensive datacenter and DTT in the data-intensive datacenter.Aiming at the data-intensive tasks and computing-intensive tasks in the datacenter,this dissertation presents respectively optimized VM static placement models,which are based on computing-intensive and data-intensive separately,so as to minimize the total task comple-tion time in the datacenter.From the view of model analysis,the proposed model computing-intensive is a linear programming problem.Therefore,we obtain the optimum solution of our model by the branch-and-bound algorithm that its time complexity is O(2NM).Simultaneously,we also present a greedy algorithm,which has O(NM)of time complexity,to solve this model.The data-intensive model is a NP-complete problem.We give a corresponding proof in this dis-sertation.Simultaneously,the dissertation gives heuristic algorithms to solve the data-intensive model.Experimental results show that the proposed method can reasonably optimize VM static placement and effectively decrease the total task completion time of datacenter.3.Improve the datacenter's total throughput of the execution tasks based on the method of combining dynamic placement and static placement in the datacenter.The most of pervious works have primarily focused on dynamic placement or static placement separately.There are few work to have been attention on jointly taking dynamic placement and static placement into account.Existing studies show that only consider the VM dynamic placement or only consider VM static placement cannot have satisfied the datacenter which has a large number of VMs processing task requirements.However,a good VM placement strategy will statically place as many VMs as possible into the datacenter.At the same time,with the change of usage of PM resources in the datacenter,VMs continuously migrate between different PMs,which can improve the total throughput of the execution tasks in the datacenter.The proposed optimization model in this dissertation combines the VM static placement and the VM dynamic placement by time slots,and minimizes the cost of migration in the process of dynamic placement.The simulation results in this dissertation show that our approach can improve the total throughput of the execution tasks in the datacenter and minimize the cost of migration.