Auto Shift: Energy-Aware Server Provisioning
Green, as we now know, is, indisputedly, the new black. Seems like you can't turn on the television or pick up a newspaper to read about the latest green initiative. Lots of people are talking.
Feng Zhao is doing something about it.
Zhao's Networked Embedded Computing group is showing a TechFest demo called Auto-Shift: Energy-Aware Server Provisioning, which addresses server resource management for Internet services, such as Live Messenger and Hotmail. Data centers for such services require potentially expensive decisions about how many computers to allocate and how those are deployed.
"No. 1," Zhao says, "you have to buy the servers. No. 2, once you buy a server, you have to manage it. And third, you have to have an infrastructure, such as power supply. In this particular study, we looked at the power usage of the servers that are running one of our largest Web services. If you look at the load as it varies over the course of the day, it peaks around noon and slows down around midnight. That clearly shows that not all the servers are needed all the time. Can we shut down some of the servers? Can we actually save energy?"
This demo is the same paper I referred to earlier in another post.
The blog entry continues with the following points made by Feng.
"We also have all these sensors in the data centers," Zhao says. "Some of the machines work harder than others. If we can move the workload around, from hotspots to cool spots, the air conditioning doesn’t have to work as hard, because of the efficiency of cooling the hottest spots. If you move that workload and even out the temperature disparities, that means good energy savings. Incorporating environmental-sensor readings such as temperature and humidity, and couple that with smart scheduling and workload migration, and we believe we can even save more resources."
That sounds green, indeed--and economical, too.
"What it translates to," Zhao concludes, "is that you use less power and that, with these smarts, we can figure out that maybe we don’t need to buy that many machines to start with, because we can do the same work, with very little difference in performance, and actually run it on a smaller set of machines. Reduce energy cost and reduce hardware investment in the first place--that would reduce service cost, reduce staffing, and reduce the space you need to build."