Energy Management Technology Research In Virtualized Data Center

With the appearance of Web2.0and popular of social network web-sites such as Facebook, etc. which makes the contents of internet exploded in the past several years, also lead to the second round of intensive con-structions of data centers all around the world. Data centers have become the necessary infrastructure as power plant in our daily life, which not only provide the storage and dissemination of information, but also exist as an commerical product under the driving of cloud computing. As a high en-ergy cost entity, Reducing the energy consumption of data centers could help the operator save the operation cost, which has obviouse economic benifits; also could cut down the emission of carbon dioxide, which shows the benifits for the society; the last benifit is for the government, who can put its promise into practice, which will help the government lessen criti-cize from international society.These research work of this paper is finished under the above motiva-tions and background, we extend our research on the energy management of data centers from three different levels namely data center level, chassis level and application system level. Correpondly, we solved three different problems in current data centers, the first is the unbalanced thermal dis-tribution in current data center, which may lead to the increase of cooling energy cost; the second problem is the different energy efficiency between the two power domains in the chassis, which will cause the increase of computing power potentially; the third problem is the wrong direction in previous research, which mainly focused on the energy saving in data cen-ter, but ignored the influence to application performance running in data centers, we need to correct such neglectness to the application performance in energy management of data centers.This paper mainly has following innovations:·A thermal-aware workload placement policy:In order to reduce the energy waste brought by the unbalanced thermal distribution in the data center, we build a model based on the neural network and rein-forcement learning method, which has learnd the history data records and evolved from every past sample data record, then build the rela-tionship between the workload and inlet temperature. This model is then used to predict the inlet output if the incoming workload is dis-patched to some chassis in the rack, we then chose the best inlet output to decide which chassis is the most appropriate destination for the in-coming workload, which will help lessen the cooling energy waste in data center.·A current-aware capacity managemet policy:this method is proposed when we found the different power efficiency between the two power domains in the chassis. We observed that the bigger current differ-ence between these two power domains, the higher energy cost of the whole chassis, so we try to utilize such difference to save the com-puting energy cost of whole chassis. We analyzed the experiment data we got to find out the relationship between workload and elec-tric current for these two power domains, then a proportion-intergral controller is implemented to monitor and adjust the workload place-ment and resource allocation between these power domains, which could minimize the current difference between these two power do- mains to keep the whole chassis in the most energy efficiency state to save computing energy.·Dynamic configuration of3-tier web application system based on the workload prediction:We make use of the easy power on/off feature of virtual machine to implement the scalability of web application sys-tem in cloud computing environment. At the beginning, we found out the relationship between user capacity and system configuration through a series of experiments, then we analyzed the histroy request record to find out the periodic characteristic of the daily requests and the required system configuration for the requests in each time period, which is an advisor for the dynamic change of the system configura-tion. This method not only guarantee the quality of the application, but also save the resource and energy consumption in data center.·Coordinated cyber-physical system for performance and energy man-agement in data center:To our best knowledge, we are the first to use the cyber-physical system in enterprise data center energy man-agement. We proposed a balance of payments model to coordinate the application system control in cyber world and the energy man-agement system in physical world. This model is used to judge every command from cyber and physical control system, if the execution of the command could lead to positive value of the model, then it was accepted, otherwise, the command will be neglected. The objective of the model is to maximize the benefits of the data center operators, of course, this objective should be achieved under the prerequisite of satifying the SLA. At the end, we validated the efficiency and ben-efits brought by the cyber-physical system through the experiments compared to cyber and physical control system seperately. All the experiments in this paper were finished in the mid-size data center provided by the CERCS lab in Georgia Institute of Technology, part of the data was got from algorithm simulation in MATLAB, most data was produced from the real experiments in the data center. We use the ESXi server and vCenter server and client provided by the VMware company. All these experiments results have indicated that energy management in data center especially the cloud data center will require the integration of cyber-physical systems to increase the interaction between cyber world and physcial environment, which will look after both sides between customers and operators. CPS will help manage the application performance and en-ergy consumption from holistic perspective to achieve more profits and save more energy.