Quantum and high-performance computers are ultra powerful, but their massive energy usage is a climate challenge. Researchers at Northeastern have developed a slew of cutting-edge solutions to help.
Quantum and high-performance supercomputers are some of the most powerful tools we have for solving some of the world’s most pressing problems, but they come with a cost: energy usage.
Sustainability has become one of the most pressing issues in the world of ultra-powerful computing. It wasn’t always the case. For a long time, the industry failed to recognize what some see as its darkest secret. But that has changed in recent years, in part due to the work being done by a team of researchers at Northeastern University who have been championing the need for more sustainable computing even before the industry accepted it as the necessary next step.
“Sustainability is something that will become more and more important,” says Devesh Tiwari, an associate professor of electrical and computer engineering at Northeastern and director of the Goodwill Computing Lab. “We really have to think about the next generation. … People are starting to think about it, and my students are actually proposing solutions to these problems.”
For years, Tiwari and a team of students have been developing cutting-edge solutions to the energy usage question, work that was honored at the most recent International Conference for High Performance Computing, Networking, Storage and Analysis, otherwise known as SC (for supercomputing. Four of the 99 papers selected from about 470 submissions at the conference, which had 18,000 attendees, were from Tiwari and his students. Two of them were selected as best student paper award finalists out of a total five finalists. The norm is for a researcher to have one or two papers selected in their lifetime.
Tiwari’s group has an exceptional record stretching back years. In 2023, his group had four accepted papers at SC out of 90 accepted papers, while in 2022 and 2021, his group had two papers accepted.
“We have been really creative in staying ahead of the curve, not just solving old problems and making incremental progress,” Tiwari says.
Tiwari says the conference also highlighted how work is not just about quantum computers but people. His former PhD stude Tirthak Patel, who is now a tenure-track assistant professor at Rice University, was there to receive the ACM SIGHPC Outstanding Doctoral Dissertation Award.
“Seeing people whom you have mentored, achieve recognition and do exceptionally creative things is the biggest achievement and best part of my job,” Tiwari says.
One of Tiwari’s Ph.D. students, Yankai Jiang, developed an innovative way to reduce the carbon footprint of serverless computing, a form of cloud computing used by companies like Google.
The challenge Jiang was trying to tackle was a simple but necessary one.
“The user always prefers the fast hardware and the new hardware,” Jiang says. “The reason people select this hardware is because they want to save time. They want to get results really fast, but the problem is it will consume lots of energy. It’s not energy efficient and it’s also not carbon efficient.”
Jiang’s research, which is some of the first to lay out a method for reducing the carbon footprint of serverless computing models, proposes a method called Ecolife. It involves using a combination of new and old hardware to reduce carbon emissions while maintaining a certain level of performance.
There are trade-offs when using old and new hardware that Jiang acknowledges can be balanced when using a combination of the two. Older hardware works slower and generates more carbon when executing code or a computing function, but it requires less carbon to manufacture. New hardware has higher performance and generates little carbon generation when running, because it’s faster, but much more carbon during manufacturing.
Jiang proposes integrating old and new hardware in serverless computing by switching between them, computing for a short time on the new hardware to maximize its benefits before switching to older hardware for a longer period of time. When combined in Ecolife, old and new hardware work in concert to approach the energy efficiency and performance of some of the best computing models out there.
“So, don’t drop your old hardware,” Jiang says. “Sometimes old hardware is more useful compared to new hardware.”
Another of Tiwari’s Ph.D. students, Ana Luisa Solórzano, helped deploy an incentive-based program that encouraged users of one of Japan’s top supercomputers, RIKEN’s Fugaku, to cut down on their energy consumption.
Tiwari’s group had started working on incentive-based programs for HPC systems in 2020, but this is the first real-world deployment.
The RIKEN incentive-program, termed Fugaku points, offered users of the supercomputer three energy saving methods they could opt to use when running their programs. The people who opted to use these energy saving methods, at possible performance expense could then spend those points to access a priority queue that would essentially let them jump ahead in the long line to access the supercomputer.
The program was implemented in two rounds where people could accrue and then spend their points. What Solórzano found was that users were incredibly intentional about how they used the program.
“[While] they were accessing the priority queue, they were not as worried about the power savings because they were taking advantage of the benefit to run the applications and get best performance,” Solórzano says. “At the same time, the same users were running other applications without the priority queue but being responsible.”
The program ultimately saved 10.7% energy in the first round and 13.26% in the second round.
“Incentive-based [programs] are the future because more and more the responsibility [of sustainability] needs to be shared not only for administrators but users,” Solórzano says.
While sustainability remains a pressing issue, Tiwari adds that his team is doing work that is also breaking new ground in areas like natural language processing on quantum computers. Daniel Silver, a Ph.D. student in quantum machine learning, developed the first real demonstration of an IBM quantum computer recognizing, classifying and understanding human language used in texts like movie reviews.
“I always tell my students, it’s about coming up with the simplest and most creative solutions that entirely change peoples’ minds about things and their perspective,” Tiwari says. “Changing people’s perspective slowly and positively is one of the most satisfying impacts one can have.”