|
Greening the Internet with Algorithms
There may be plenty of free things associated with the Internet, but energy isn't one of them, as anyone who runs a data center knows. Amazon, Google, and other firms that use multiple data centers for handling requests over the Internet pay tens of millions of dollars to keep all that power going. Now, thanks to some work by researchers at MIT, Carnegie Mellon University, and networking company Akamai, such companies may be able to slash their energy bill by 40 percent.
The trick is to reroute data to locations where electricity prices are lowest on a particular day or at a particular time. Asfandyar Quireshi, a PhD student at MIT, proposed the idea in a paper. He and his colleagues then approached Akamai to get the real-world routing data to test his theory. They also collected and analyzed 39 months of electricity price data for 29 major US cities. Because prices can change for a variety of reasons, the researchers found huge variations, even among geographically close locations, with no one place that was always cheapest.
The scientists came up with an algorithm to take advantage of daily and hourly price fluctuations, which would balance such factors as the physical distance needed to move the data (longer distances cost more) against the cost savings from the reduced energy use. They then tested the algorithms using data from nine Akamai servers covering 24 days of activity. The best scenario showed a reduction in energy consumption of 40 percent. Not every company would be able to manage that especially given that many servers do not consume significantly less energy when they are idle than when they are running at full blast but it may be a start. And when you're talking about that much money, every percentage point adds up, especially since some estimate that data center energy usage will quadruple in the next decade without some effort to improve efficiency.
Read
more about this
Cell Phones Used to Track Relationships
Your cell phone just might know who your friends are better than you do. That's one way to read the results of a study conducted by researchers at the SatanFe Institute of New Mexico, the Northeastern University in Boston, and MIT. They handed out nearly 100 cell phones to volunteers at MIT, who signed comprehensive consent forms because the phones contained software that logged users' calls and could detect when another of the phones in the study was close by.
The phones could predict with 95 percent accuracy which people the volunteers would identify as their friends. Not only that, but the researchers could use the data to predict who would report more or less satisfaction in their jobs; those who were less satisfied didn't tend to have friends in close proximity at work and were more likely to call friends during work hours. Interestingly, judging from the cell phone proximity data, people tend to overestimate how much time they spend with close friends and underestimate how much time they spend with casual contacts.
So how could this data be used? Certain kinds of studies might be able to use software on personal gadgets to measure behavior rather than having the researchers conduct interviews, which take a lot of time and effort and can be unreliable. Proximity studies may help epidemiologists predict how contagious diseases spread. And marketers have already started taking advantage of this potential goldmine: Teradata has helped a Canadian network operator identify users with many connections to be targets of refer-a-friend offers. So far, this approach performs better than more conventional marketing campaigns. However information obtained from our gadgets is used, though, privacy issues must be taken into consideration, possibly by using techniques to make the data as anonymous as possible.
Read
more about this
Going Modular for New Motherboard Design
What does it take to create a new computing paradigm? Sometimes it calls for thinking outside the box; other times, it calls for thinking within the box lots of boxes. One group of researchers is doing just that. Their new prototype for a motherboard takes the processor, memory, and storage, and places it on a tiny module. A PC would use lots of these modules arranged in a networked cluster in which the whole might be superior to the sum of the parts.
The design, dubbed the Illuminato X Machina, delivers a CPU, RAM, data storage and serial ports for connectivity on every two square inches, explained project contributor David Ackley, associate professor of computer science at the University of New Mexico and one of the contributors to the project. One key advantage of such a design could be a decreased frequency of system crashes. If a single cell fails, the rest of the system could remain operational. Also, project designers think their modular design could lead to machines that draw very little power.
Each X Machina boasts a 72 MHz processor, a 16 KB solid stat drive and 128 KB of storage. It can connect to its neighbors on all of its four edges, and it knows when it's plugged into another X Machina so it can establish the proper connections for exchanging power and information. While it's a richly detailed design, it hasn't been benchmarked yet, so those working on the project don't know how a computer powered by a cluster of X Machina modules measures up against a more conventional PC running, for example, an Intel Core 2 Duo chip. But Ackley's students will get their chance to see what the design can do next month.
Read
more about this
|
