“Energy efficiency” is the current trend. Cars are typically ranked by fuel consumption per mile, not top speed. Consumers select light bulbs with the electric bill in mind. Everything, it seems, is going green.
The world’s top supercomputers are now following that path thanks in part to the work of Wu Feng and Kirk W. Cameron of the Virginia Tech College of Engineering. With a staff of volunteers, Feng and Cameron have been issuing the Green500 List twice a year since 2007. The Green500 ranks supercomputers by their energy efficiency, serving as a complement to the better-known Top500 List, the industry-benchmark ranker of powerhouse supercomputers.
“A lot of people who tended to scoff at green computing have embraced it,” said Feng, associate professor with the Department of Computer Science and the Bradley Department of Electrical and Computer Engineering.
“We initially started the Green500 List to raise awareness and encourage reporting because you can’t optimize what you don’t measure,” said Cameron, associate professor of computer science.
The reason for the Green500’s growing clout is simple: Power is the life supply and kryptonite to supercomputers. With thousands of computer nodes hooked together and producing great quantities of heat that must be cooled by giant air conditioners, these behemoths require a great amount of electricity to operate. Case in point: The current No. 1 supercomputer, located in China, uses 4 megawatts and ranks high (No. 11) on the most recent Green500 List for its conservation efforts.
Future supercomputers could consume about 100 megawatts, thus requiring their own power stations. Eventually, a top-notch supercomputer could require 1 billion watts to operate if designers do not improve their energy efficiencies. In comparison, the entirety of New York City uses roughly 5 billion watts in a year.
Kirk W. Cameron, a co-founder of the Green500 List, also helped build Virginia Tech’s $1.2 million System G (for “green”) supercomputer. It is made up of 325 Mac Pro computers, each with two four-core, 2.8-gigahertz Intel Xeon processors and 8 gigabytes of random access memory, cranking out 22.8 teraflops (trillion operations per second) of computing power.
After the release of the fall 2010 Green500 list, trade magazine HPC Wire reported: “The need to maximize energy efficiency in compute-intensive environments has never been greater.” On its website, FastCompany reported: “Before long, the fastest supercomputers may also be the most efficient.”
That wasn’t always the case. While working for the federal government in 2001, Feng proposed a project called Supercomputing in Small Spaces, basically self-built, energy-smart supercomputers. Think a Toyota Prius versus a souped-up sports car. Both will get you where you want to go, but one sips gas. The other guzzles it.
Feng’s first try was a 24-node Bladed Beowulf cluster, dubbed MetaBlade. It used fewer than 400 watts of power and occupied the same space of three stacked pizza boxes. A 240-node cluster followed with a similar footprint that used only 3.2 kilowatts -- the equivalent of two hairdryers -- but delivered 100 gigaflops, or 1 billion floating point operations per second. Its name: Green Destiny, after a mystical sword in the film “Crouching Tiger, Hidden Dragon.”
Critics scoffed. One of the crueler comments Feng remembers is “Green Destiny is so low powered it runs just as fast as when it is unplugged as it does when it’s on.” But Feng won people over. “The mantra was always bigger, better, faster,” he said. “What I was proposing was in the other direction: Performance efficiency. Speed is important. But performance related to other metrics is just as important.”
Green Destiny eventually led to the Green500, which debuted in 2007 at a supercomputing conference in Nevada. Three years later, Feng calls the Green500 “a work in progress.”
Two computer science doctoral students and two support personnel from the university help Feng compile the list. Supercomputer owners submit much of the data used to compile the list through an online entry form. Many facilities don’t submit data. For those, the group culls information from machine specifications and estimates the power usage of each model.
From left: Wu Feng, associate professor of computer science and electrical engineering at Virginia Tech, heads the twice annual Green500 List with volunteer support staff help from Tom Scogland of Lake Forest, Ill., and Balaji Subramaniam of Chennai, India. Both are doctoral students in computer science. Also on the Green500 team are Virginia Tech’s Jason Lockhart, an operations support technician at InnovationSpace, and Mark Gardner, a network research manager.
“The Green500 list will play a major role in the evolution of supercomputing by keeping the power consumption and energy efficiency of supercomputers in check,” said Balaji Subramaniam, a doctoral student in computer science from Chennai, India.
Tom Scogland, also a doctoral student in computer science from Lake Forest, Ill., added: “The more visible efficiency is to the community, the more pressure we'll have to affect change.”
There are other factors for more attention toward green supercomputing, Feng said. Among those are current federal efforts to seek out green energy sources such as wind turbines and the like. Wherever the attention comes from, he said he welcomes it.
“We want to raise awareness for energy efficiency,” Feng said. “There’s no other list like this.”
Printer-friendly version
Send by email