Programming Protocol Cuts Computer Energy Use

Luis Ceze (Univ. of Washington)

Researchers at University of Washington in Seattle have created a software programming protocol called EnerJ that reduces energy consumption in simulations by up to 50 percent. The computer scientists and engineers will present their research next week in San Jose, California at the Programming Language Design and Implementation annual meeting.

The team led by UW computer science professor Luis Ceze (pictured left) takes advantage of processes that can tolerate minor deviations from peak equipment or software performance, without sacrificing delivery of a quality product or service. For example, image recognition software works even if the scanner screen encounters a speck of dust. EnerJ applies that principle to a wider range of computing operations, where tolerance of minor errors can lead to lower energy consumption.

EnerJ is an extension for the Java programming language that creates two interlocking pieces of code. One piece is the precise representation of data, while the other piece is a more fault-tolerant version for processes that could survive a few misses. EnerJ creates a barrier between the two pieces to prevent leakage from the approximate to the precise part.

Where systems can tolerate an approximate representation of the data — for example, a number rounded to a lower number of decimal places — the system would need less computing power to process the data and thus use less energy. Where data must be precise, as in many health care applications, the system would process that precise piece of data.

EnerJ could lead to not only more efficient use of energy in computing, but also enable the placement of more transistors, each using less power, on a single chip. Simulations of hardware running software with EnerJ show energy could be cut by 20 to 25 percent. One program recorded a simulated energy savings of almost 50 percent.

From redesign of software alone, the researchers estimate EnerJ could result in energy savings of 30 to 50 percent. Researchers are now designing hardware to test their results in the lab.

Read more: Engineers Increase Computer Program Speed, Retain Safety

*     *     *