Purdue University

It’s a problem familiar to every college student – courses are oversubscribed, and you can’t get into a class you need to stay on track – or even worse, to graduate. Now a Purdue professor is using Rosen Center for Advanced Computing (RCAC) supercomputers to study this problem.

Kevin Mumford, a professor of economics, is using the Bell Community Cluster to simulate thousands of course registration outcomes. His goal is to analyze how Purdue’s batch registration system impacts students, particularly when they are shut out of essential courses.

By using the Bell cluster, Mumford can conduct large-scale simulations that would be impossible with standard computing methods.

With these large-scale simulations, Mumford can quantify the probability of a course shutout for each course requested by each student and estimate how this affects their educational path, particularly in STEM fields.

Mumford’s research focuses on Purdue’s batch registration system, which assigns students to courses based on preferences. The conditional randomness in the registration system makes it possible to use the simulated probability of course shutout to estimate the causal effect of not being assigned to a requested course. “It’s like a little quasi-experiment,” explains Mumford.

Running these simulations requires significant computational power, as each one takes about 12 hours to complete. To manage this, Mumford relies on the Bell cluster. Initially, he and his team used a computer lab to run the simulations, but this method proved inefficient. “We took over a computer lab that was not being used during the summer and ran the simulation on every computer in the lab,” Mumford says. “But there were some issues, and that process took time every day.”

The Bell cluster, with its high-performance computing capabilities, allows Mumford to automate the simulation process, running many simulations simultaneously. When each simulation is completed, there is no downtime before the next one begins. This has dramatically sped up the research process. The ability to run such large-scale simulations would not be feasible without the Bell cluster.

“If I was running this on my desktop computer ... it’s 19,000 simulations times 12 hours each, and that’s a lot of computer hours,” says Mumford. “What Bell does it automate this process for me and make it run much faster.”

So far, Mumford has completed about 8,000 of the 19,000 planned simulations, starting with student course preference data from the Fall 2018 semester. The results will provide insights into the likelihood of students being shut out of specific courses and how this impacts their academic decisions. He is particularly interested in how these shutouts affect students’ long-term educational trajectories.

“Getting shut out of calculus is a really big deal,” says Mumford. “It significantly decreases the probability that a student’s going to graduate in anything STEM-related. It is important that there be sufficient capacity for students to enroll in certain critical courses.”

He believes his findings could inform future academic policies, particularly around course availability and registration, emphasizing the need for access to key introductory courses early in students’ academic careers.

To learn more about the Bell Cluster and other RCAC resources, contact rcac-help@purdue.edu.