I just got an email regarding a notification of paper acceptance. Well, our new paper “Adaptive Power Management for Data Center in Smart Grid Environment” has been accepted to be published in the proceedings of the conference The 10th IEEE International Symposium on Parallel and Distributed Processing with Applications (ISPA-12).
The authors of the paper are  Rakpong Kaewpuang (the main contributor of this paper),  Sivadon Chaisiri (me)  Asst. Prof. Dusit Niyato,  Assoc. Prof. Bu-Sung Lee, and  Asst. Prof. Ping Wang. We are from School of Computer Engineering (SCE), Nanyang Technological University, Singapore. One of the authors will present the paper in this conference. This IEEE conference will be held in Universidad Carlos III de Madrid, Spain between July 10-13, 2012.
Let me introduce a bit about the paper as follows:
In the paper, we propose a mathematical optimization model to achieve the optimal solution for the power management of data centers. This power management optimization is preliminary derived for a data center based on smart grid technology. Suppose that the data center has a built-in renewable electrical power source (e.g., wind turbine and solar panel). In the real world, some data centers have already applied these kinds of renewable power sources.
Well! it is a very nice idea to utilize such a greener energy source. The data center can generate an amount of power to operate its own facilities. However, the data center cannot rely on only the power generation of the renewable source since the generated power might not be sufficient to run the data center. Why? ….. The main reason is that the power generation of a renewable source strongly depends on the natural environment. For example, the power generation of a solar panel depends on the strength of sun light while that of a wind turbine depends on the strength of wind. That’s right! we cannot predict the nature. Someday it’s cloudy and someday it’s not windy. Don’t forget that a data center is an energy hunger!! What will happen when the cloud is dark while the solar panel will take leave for a few days?
So the data center still needs to purchase an additional amount of power from the electrical power grid. Let’s say that the smart grid is available in the city where this data center is located. (* Some countries have already applied the smart grid technology.) Then, power prices based on demand-and-supply dynamically vary. We may call such a fluctuating price “spot price”. So sometime we can luckily purchase the power from the electrical grid with a very cheap spot price. However, sometime we may purchase the power with a very expensive spot price.
Now we have two kinds of uncertainty, i.e., 1. power generation of renewable power sources 2. spot prices in smart grid. Actually, we still have another major uncertainty, namely the computational demand uncertainty.
Let’s talk about this demand uncertainty. The demand uncertainty is about the unpredictable number of transactions that accesses the data center. The data center (especially where supplies cloud computing services) encounters the great energy cost. The large portion of operation cots is from the amount of power required to operate the data centers (e.g., a large number of servers and cooling devices). In particular, the energy cost also directly depends on the demand (uncertainty). That is, a data center may not need to turn on all servers since some servers may not be fully utilized (in fact, some server could be mostly idle). So the power consumption could greatly decrease when a number of servers can be turned off (or be turned to the standby mode) due to the lower demand volume.
In this paper, we propose the optimization model based on stochastic programming with multi-stage recourse to achieve the optimal solution for the power management. The aforementioned uncertainties are mainly addressed. We also recommend that a power storage (i.e., battery) can be applied to keep the power generated by the renewable source, and also store the power purchased from the electrical grid when a spot power price would be considerably cheap. Again and again, the uncertainties of power generation, spot prices, and demand are involved. Hence, it is a major challenge for data centers to achieve the optimal under uncertainties. That is, it is not trivial to know how much the amount of power we do need to purchase in advance from the electrical grid and how much the amount of power we do need to store in the battery. Here, our contribution is the the optimization model which can deal with such a problem.