Design Of Virtual Data Center Power Consumption Management Adaptive System

With the development of cloud computing and virtualization, thepower consumption management of data center abstracts more and moreattention when the application-level quality of service(QoS) is needed to beensured at the same time. However, due to the complexity of theapplication and dynamic workload, power consumption reduction of datacenter has to be further studied to ensure the QoS and the adaptive resourceallocation of data center.In this paper, we analyze the existing power consumption managementtechnologies and methods of virtual data center and propose a powerconsumption management adaptive control system which is based onintelligent control theory. The adaptive system consists of resource monitor,resource coordinator and resource allocator. The main innovations of thispaper include the following three points:1. The system is able to implement power consumption managementadaptively, that is to say, dynamic resource allocation can be carried outaccording to time-varying workload. Online estimation accuracy of powerconsumption and performance is the foundation of power consumptionadaptive management system. Resource monitor is responsible formonitoring overall power consumption of all the physical servers and theperformance of each virtual machine, the results are sent back to modelpredicator. Model predictor is charge of online identification about themodel which represents the relationship between the physical cluster powerconsumption and virtual machine CPU throughput using particle swarmoptimization algorithm. The identification results are sent back to the resource optimizer.2. The system can make a trade-off between power consumption andperformance. Power consumption can be reduced under the premise ofmeeting data center QoS. The power consumption reduction is finished byresource optimizer. First, utility function about power consumption andperformance is optimized according to the prediction model using geneticalgorithm. Second, optimization results are assigned to each virtualmachine. Third, optimization resource allocation is executed by XenVirtual Machine Manager to balance power consumption and performancemanagement of virtual data center.3. The system has a great scalability. Particle swarm optimizationalgorithm and genetic algorithm are used for the model building andoptimization. Because there is no need to consider the specificcharacteristics of the model by intelligent control theory, this method is aptto the complex nonlinear system and model. Meanwhile, multipleparameters can be set. It is easy to apply for the more constrained model inthe implementation. In this way, scalability of this adaptive system isimproved.To verify the effectiveness and stability of this system, test platformbased on Xen is set up using the TPC-W as workload generator. Throughthe analysis of the experimental results, the system is not only able toguarantee application-level QoS but also to allocate virtual machineresources adaptively. Besides, the adaptive system may bring energysavings to some extent.