| With the development of virtualization technology, the cloud computing platform based on virtualization technology gains a lot of attention. More and more fields use cloud platform to perform their jobs. Cloud platform virtualizes and encapsulates all physical resources, and provides them to users in the form of Virtual Machine. By using virtualization technology, cloud platform not only divides and schedule physical resource in a more fine-grained mode, but also be able to dynamic manage resource. Typically mappings between applications and physical nodes affect the cloud platform’s throughput, load balancing, and energy consumption. The elastic characteristic of cloud platform makes the online adjustment of resources allocation is possibleNow, the problem of huge energy consumption is caused by cloud platform becomes hot. Large scale physical clusters are used by cloud platform to run jobs. Meanwhile, lots of physical nodes are active because of underutilized physical nodes exist. This is one of the main reasons that lead to the extremely high energy consumption. Large number of underutilized physical nodes has become a critical problem to cloud providers that needed to be figured out. Some researches focus on the technology of virtual machine live migration to improve the energy efficiency of cloud platform.The technology of cloud computing is increasingly developing, and application types become diversification, such as the parallel computing and other distributed applications. The virtualization technology in the single machine environment cannot meet the requirement of parallel computing applications. Distributed virtualization technology is developed and gains more attention. Virtual Cluster is proposed based on the development of technology of Virtual Machine and Virtual Network. Cloud platforms use virtual cluster to provide a high performance computing environment which is customized, independent, and reliable for parallel and distributed applications. However, due to a lot of virtual clusters are used in cloud platform, the management of energy-efficiency is become complicated:(1) Virtual cluster is composed by a group of correlative virtual machines. For the energy consumption management of the cloud platform which virtual clusters exist, we cannot use the technology of virtual machine live migration because it cannot guarantee the correctly running of applications. So when modeling the energy-efficiency problem, researchers need to consider the relevance of virtual machines in certain virtual clusters.(2) Based on the character of parallel application program, virtual machines in the same virtual cluster are usually created and revoked at the same time. When virtual cluster is mapping on multiple physical nodes, the energy efficiencies of these physical nodes will be affected by the virtual cluster, and more communication costs will be generated.Besides, the utilization of memory becomes the main reason for how many jobs can be executed by a data center. Memory is more critical to data intensive job. If the allocation of memory for virtual machines is not reasonable will generate the resource fragments and influence the utilization of physical node.Focus on above problem, we use virtual cluster live migration technology to consolidate virtualized resources. This technology migrate the whole environment of parallel jobs and ensure the execution of jobs are efficient and correct. After analyzing how the placements of virtual clusters have an effect on the running energy-efficiency of cloud platform, we use the remaining time of virtual machine as a factor to evaluate which virtual machine should to be migrated and which physical node is proper to hold the migrated virtual machine.In order to realize the high energy-efficiency, we formulate this object into a combinational optimization problem that improving the amount of data which is processed by unit energy consumption under the performance constrains. We proposed virtual cluster live migration schemes to solve the virtualized resource consolidation problem.Firstly, we only consolidate virtual machines based on their memory requirements. We formulate this kind of problem, and a two stage consolidation scheme is proposed. In the first stage, we use the correlation of virtual machines belonging to same virtual cluster and whether virtual clusters’ remaining running time is similar to divide all virtual machines into several groups. The VCGM algorithm is designed to construct a final mapping state which has high energy efficiency. In the second stage, the low cost perfect machine model is used to match physical nodes in final mapping state and initial mapping state in order to generate a migration plan.Secondly, in order to fit the more complicate and practical data intensive job model, we consider the resource of memory, CPU and I/O together. We also formulate this multi-dimension virtualized resource consolidation problem, and proposed a virtual cluster live migration scheme which is based on the ant colony optimization algorithm, called ACVCM. We define the benefit reward and make it as a component to compute the heuristic information. We hope to generate a migration plan which balances the two purposes of reducing the difference of remaining running time of virtual machines mapping on each physical node and reducing the total migration cost.At last, in order to verify the effectiveness of both migration schemes, we have made some simulated experiments to evaluate them. We also discuss about the experiment result carefully. The two stage migration strategy reduced the energy consumption by 42%-55%, and ACVCM reduced the energy consumption by 37%-42%. These data illustrate that the proposed virtual cluster live migration strategy can effectively improve the energy-efficiency of cloud platform. |
PDF Full Text Request