Carnegie Mellon's Pulkit Grover Developing Tools To Reduce Energy Consumption in Data Center Communication Networks

We are intoxicated by technology. We are seduced by its power, its speed, its gadgetry and its promise to solve the problems of the world. Technology is a carrot we have trotted after for years, and as it speeds up we gallop after it.


But that speed requires energy. So as we speed up technology, we create a new class of energy gluttons, who are predicted to consume more than 15 percent of the world's electricity to power today's information and communication technologies.


Carnegie Mellon University's Pulkit Grover, an assistant professor of electrical and computer engineering, is developing tools that will help reduce the amount of energy consumed by data center networks by 30 to 50 percent.


"It is becoming difficult to reduce transmission energy or circuit energy in isolation. We are finding that there is significant room for improvement at the intersection of circuits and communication. "These networks usually consume 10 to 20 percent, and even up to 50 percent of the data center energy," Grover said.


Grover, who is scheduled to present his research at the July 7-13 International Symposium on Information Theory in Istanbul, Turkey, reports that his team is developing new energy-efficient protocols and circuits that can reduce energy consumption by including the right patterns of "redundancy" for correcting errors in communication links of big data centers.


"Through appropriate use of redundancy, we can correct errors with significantly lower total energy use compared with current systems," Grover. "We incorporated power consumed in transmitter/receiver circuitry in Shannon's information theory, and obtained new fundamental limits on total power usage."


The solution is timely: companies are currently standardizing data center communication protocols. "My goal is to influence the standardization process so that energy efficiency is addressed meaningfully, and to engage with industry on many levels. Networking in data centers is just one example of where these techniques are applicable," Grover said.


In the past two years, this research has been supported by the Interconnect Focus Center of the Semiconductor Research Corporation, and by the National Science Foundation (NSF) Center for Science of Information. It is currently supported by a seed grant from CMU, and a travel-related seed grant from the NSF Center for Science of Information.