Apr 22

Energy-aware traffic engineering

Hello ecebloggers,

Yesterday, Tuesday 04/21/2009, a close friend of mine, Nedeljko Vasic (in the picture below), gave a talk at UNH CS Weekly session on his current research titled Energy-aware traffic engineering. Nedeljko is currently a second year PhD student at EPFL’s Networked Systems Laboratory. In 2006 he was awarded with “St. Sava’s” award by the Ministry of Education of The Republic of Serbia for being the best student in the country. He also received IBM PhD Fellowship for 2009.

What is energy-aware traffic engineering? As Nedeljko mentioned in his talk, Internet’s energy consumption in the US only is 20TWh which costs about 2 billion dollars per year. With demands like cloud computing, video streaming, and video on demand energy consumption is highly likely to increase. CMOS technology is reaching a plateau in power efficiency and cooling might help, but actually coolers will just increase the energy consumption. Nedeljko is suggesting that the solution for this problem lies in a protocol that enables network hardware (routers, network cards …) to be smarter and aware of the amount of energy it needs to optimally address the network load.

Current network hardware operates in 5 different energy modes depending on the amount of network load. If the network load just slightly jumps above a certain operating region it will have to switch to a higher energy consumption mode. This switching between the energy regions is creating peaks in energy consumption, similar to turning on and off your computer. Another network device (in the network) might take this additional load and still operate in its current energy mode. Even more, network devices could take the load from several devices, stay in the same operating region, and if the devices from which the network load was taken don’t have any network load left, they could be put to sleep. By implementing this idea in EATe protocol Nedeljko was able to get within 15% of the optimal energy consumption solution (see picture below), in the worst case scenario, and remove traffic from 15-31% of the devices.

His talk raised a lot of questions in the audience and made the Q&A session 30 minutes long! One of the more interesting questions was how would load switching be provided amongst different ISPs without them revealing a lot of information about their network topology? Professors from the UNH CS department were delighted with his talk and asked him to give another one this fall. I certainly hope that he will accept their invitation:)

Have a good one,

Nemanja Memarovic