Does city-run internet service encourage private companies to invest in fiber networks early? Or does it discourage them from investing at all?
To answer this question, Associate Professor of Economics Kyle Wilson turned to Pomona College’s high-performance computing (HPC) to simulate entry and investment actions of internet service providers in 30,000 U.S. zip codes.
Models that would normally take him weeks to estimate took a day with the HPC.
“I don’t think it’s an exaggeration to say that, without the HPC, there’s no realistic way I could have met deadlines and the expectations of the reviewers,” Wilson says.
Wilson is one of many faculty members who have taken advantage of the powerful cluster of computers housed in the Cowart Building, the facility for Pomona’s Information Technology Services.
You’re giving each person the fastest, most high-quality computer for their particular research at that one moment.
“It’s not just a souped-up computer,” says José Rodriguez, vice president and chief information officer. “It’s a whole environment of computing management that can provide a foundation for higher level research.”
Locked inside a climate-controlled room, the HPC consists of multiple computers stacked on top of each other. Each “minicomputer” is attached to the others using optic fiber cable.
“The HPC is all about having the highest quality components, the fastest components, the fastest connection between components, and then scheduling for dedicated resource,” says Rodriguez. “You’re giving each person the fastest, most high-quality computer for their particular research at that one moment.”
Like Wilson, much of Professor Jo Hardin’s work relies on simulating data under different scenarios. To do that, Hardin, who is the Hardison Chair of Analytical Thinking and professor of mathematics and statistics, runs the same model repeatedly, applying one technique to different data sets. Instead of running one data set at a time, however, she can run hundreds in parallel using the HPC.
It’s made my results much more effective, allowing me to communicate my ideas in ways that are convincing to people.
“The HPC has made the process by which I do my work much more efficient,” says Hardin. “And it’s made my results much more effective, allowing me to communicate my ideas in ways that are convincing to people.”
Hardin’s student researchers have also tapped into the HPC’s capabilities, and, through the process of using the HPC, have been able to learn new skills like submitting batch jobs and using shell commands, skills they would have needed to learn in a specific computer science class.
“It’s really important for them to have this experience,” says Hardin. “It’s a very hands-on procedure.”
Yuqing Zhu, assistant professor of neuroscience, relies on the HPC for her work bridging neurobiology and machine intelligence, as do students in her Artificial and Biological Vision class. The course explores how vision works in biological systems across different animal species as well as in machine learning models.
“The students learn some advanced models,” says Zhu. “And if they’re ambitious, they can use the HPC to build and train larger models.”
Since brains have so many neurons and connections, Zhu says, larger models produce results that are more generalizable to the brain itself, whereas results in small models might not reflect how real neural processes work.
Zhu credits Andrew Wilson, director of research computing and digital scholarship, and Viet Pham, system administrator of high-performance computing, for supporting her and her students at every step.
“Andrew and Viet have been tremendously helpful in guiding us through set-up, troubleshooting and onboarding of new research students,” says Zhu.
Wilson, with his Ph.D. in computational archaeology, brings firsthand research experience to his role and is an ideal person to support faculty members in their use of the HPC.
“I know what it’s like when you have a question that’s just out of reach computationally. When faculty members come to me, often they can see where their research could go, but the tools aren’t there yet,” says Wilson. “The HPC changes that conversation.”