The use of high-performance computers has exploded. The complex systems play a role in all sorts of tasks these days, including trading stocks, forecasting the weather, and breaking codes.
But people who work with the powerful machines are encountering roadblocks. In addition to the physical limitation of how much information you can store on each chip, IC developers are hampered by power-dissipation concerns. The data centers that run the computers require a tremendous amount of energy—enough sometimes to power a small city. Where to situate the centers and how to power them are sticky issues. And, as always, there is the challenge of improving computers’ performance while making them more energy-efficient.
Figuring it all out is more than just a single computer manufacturer acting alone can handle. That’s why IEEE’s newest working group, Rebooting Computing, has set its sights on improving all aspects of high-performance computing—a “soup-to-nuts approach,” according to IEEE Senior Member Elie Track, cochair of the group along with IEEE Fellow Tom Conte, vice president of the IEEE Computer Society. Created last year, the group is sponsored by the IEEE Future Directions Committee, the organization’s R&D arm.
As Bichlien Hoang, senior program director for IEEE Future Directions, puts it, the working group will “start with a small number of activities, such as developing a Web portal and organizing workshops to engage participants, as well as creating a community for them and finding ways for the community to grow.” The group plans to launch a website to provide data, list upcoming workshops and conferences, and provide a forum for discussion, comments, and other input from IEEE members.
Once those first steps are taken the group will apply for funding from the IEEE New Initiatives Committee, Hoang says.
“Revamping computing is not something that any organization or company can undertake by itself,” Track says. “IEEE has societies and councils engaged in almost every aspect of computing, so our organization is the natural place to take on these tasks.” Track is also president of the IEEE Council on Superconductivity and a senior partner with Hypres, a superconducting electronics firm.
“The goal is to completely rethink computing, from devices to circuits to architecture and software,” he declares. “IEEE will be the catalyst to spawn new thinking.”
what's in a name?
First, a name—and function—had to be chosen. The group started off calling itself the High-Efficiency Computing Working Group, and limited itself to addressing how to reduce large computers’ power consumption. It changed its name after learning of IEEE Life Fellow Peter Denning’s Rebooting Computing project, sponsored by the U.S. National Science Foundation.
Denning, who heads the computer science department at the Naval Postgraduate School, in Monterey, Calif., is tackling issues stemming from the decline of computer science education in the United States. His effort is based on his Great Principles of Computing project, which identified the scientific theories of computing and applied them in an innovative curriculum.
Progress in computing has closely tracked IC improvements that follow Moore’s Law, in which the number of transistors on a chip doubles roughly every 18 months so that performance increases exponentially. The axiom held for roughly 50 years, but because of limitations in just how small components can be made, that kind of improvement is now leveling off, at least as far as computer clock frequency is concerned, Track says.
“There was a time when you could go from 500 megahertz to 1 gigahertz readily enough, but nowadays clocks are leveling off at around 2 or 3 GHz,” he says. Individual transistors can be pushed to run faster, but doing so for the millions found on a typical microprocessor would generate unsustainable amounts of heat.
“It’s become clear that all kinds of computers, including supercomputers, are not going to advance at nearly the rates they have in the past,” Track says. “But more computing power is needed, and we need to find ways to go beyond the existing limitations.”
power in the cloud
Some say the answer is in cloud computing, but the servers running those services are going to have to perform faster. To do so, they will need a tremendous amount of power.
Companies with large computing centers, like Google, are finding that because their cloud-computer servers consume so much power, they need to locate them where they can have water-cooling systems, Track says. “So far they have been able to manage this but at a great cost.” So computer companies are looking at both making the servers more energy-efficient and improving their cooling systems.
The goal, to be reached in stages, is to achieve exascale computers that perform as many operations per second as 50 million laptops. To this end, enhancements being investigated include significantly increasing parallel processors in multicore architectures, as well as making fundamental advances in processor and memory technology, interconnects, system architecture, and software upgrades.
There’s no dearth of ideas for improvements, but each tends to be limited to only one aspect of computing. Coming to an agreement on what to focus on is the role the Rebooting Computing Working Group has accepted.
“It’s time to rethink the entire approach to computation,” says Conte. “We have been using the same models for computation since the inception of computing. We’ve tweaked and optimized every level of the stack, but to meet today’s challenges, everything has to be on the table. This will require a serious, cross-discipline conversation among domain experts.”
at year's end
The working group is organizing a workshop to be held from 11 to 13 December in Washington, D.C. Leaders have been invited from Google, IBM, Intel, and others in industry, as well as representatives from government research agencies and academic labs. Their task: to help define the challenges and make recommendations for meeting them.
“We all have the same goals but different ways of achieving them,” Track says. “The fundamental idea is that it cannot be a narrow improvement in just one aspect; it has to be improvement across the board. The total can be more than the sum of its parts.”