Power Cost Optimization Based On Prices In Data Center(s) In Smart Grid Environment

In recent years, with the popularity of cloud computing, the needs of network develop rapidly. Power from the data center or data server has continued to grow and show a rapid growth trend. A survey reports that power cost in servers may have exceeded the cost of servers and other hardware devices. Therefore, the study that reducing the power cost becomes imminent. Meanwhile, with the development of the smart grid, more power grids start the implementation of dynamic pricing mechanisms, especially Time-of-Use and real-time pricing. The differences of dynamic pricing are not only in time but also in space. When the data center or the data server with a power storage device, such as a battery, power can be stored at low price, and be released at high price. So in this context, the study of how to use the characteristics of the electricity price to schedule power and then reduce the power cost throughout the data center or server is particularly viable and important.On the one hand, studies on data center energy consumption have yielded many important results. In fact, it is also an indirect way to reduce the power cost. There exist more mature approaches:dynamic voltage scaling, dynamic voltage and frequency scaling, dynamically adjust server status, virtual machine technology and so on. On the other hand, the development of dynamic pricing mechanism brought by the smart grid, makes the data center network loads can be migrated to be executed or saved in the low tariff phases (or locations) and then can be used in the high power tariff phases (or locations), thereby directly reducing power costs.Based on dynamic price, this article focuses on three levels (From the plain to the depth, from simple to complex):optimize the power cost of single data server, optimize the power cost of server clusters and optimize the power cost of distributed data centers. For the problem of optimizing power cost of a single data server, considering the random nature of the network load and the difference of time domain TOU, we proposed dynamic programming solution and overcome deficiency that makes the network load be deterministic for processing; For the problem of optimizing power cost of server cluster, considering to reduce the power cost by using TOU difference time-domain directly and by saving energy indirectly, we proposed the establishment of Markov decision process solutions and made a combination of previous methods of study; For the problem of optimizing power cost of distributed data centers, taking into account the limitations of space and time TOU, network bandwidth and network load difference, while ensuring quality of service requirements, we proposed optimizing the network load dispatch solution. It should be noted that the power cost we are referring to is the cost of servers’power.