Accelerating economic research at UBC with high performance computing using RONIN and AWS
Dr. Kevin Leyton-Brown and Neil Newman are computer scientists at the University of British Columbia (UBC) working at the intersection of artificial intelligence (AI) and microeconomic theory. Their research demands large-scale, high-performance computing, in episodic bursts, to run parallel simulations of complex auctions.
When Leyton-Brown and Newman began research into the computationally complex auction theory behind the 2016 United States wireless spectrum auction, their ML models required significantly more computing power than their on-premises infrastructure could provide. The UBC team turned to RONIN, an Amazon Web Services (AWS) Partner, and the virtually unlimited infrastructure of the AWS Cloud, to accelerate their time to answers and new discoveries.
Powering a high-stakes government auction with high performance computing
In 2016, the United States Federal Communication Commission (FCC) wanted to reallocate radio airwaves from TV broadcasters to wireless carriers. Known as Auction 1001, this reallocation aimed to increase the amount of radio spectrum available for wireless 5G broadband services. Radio spectrum is the range of frequencies that wireless signals travel over, like lanes on a highway—and this spectrum is limited. The FCC’s goal was to make more frequencies available for wireless phone and internet services, which would help make those services faster, more accessible, and more reliable.
A team of researchers and computer scientists, including Dr. Kevin Leyton-Brown and Neil Newman from UBC, helped the FCC create an incentive auction to pay TV broadcasters to relinquish their airwaves for a price, and then sell these airwaves to wireless carriers. But designing this auction was no simple feat. Approximately 1,000 television broadcast stations submitted competing, confidential bids to sell their radio spectrum to the FCC, and wireless carriers submitted competing bids to buy blocks of “contiguous spectrum,” or a range of radio frequencies that are adjacent to each other with no gaps or broadcast interruptions.
Running the auction required solving many complex “graph-coloring problems.” In its most basic form, a graph coloring problem involves figuring out how to color countries in a map so that no two adjacent areas have the same color. In the case of the FCC’s incentive auction, the “countries” were geographic regions of the United States and the “colors” were TV channels on which stations would receive a broadcast license. The team of computer scientists needed to determine the optimal allocation of channels to stations that avoided interference between them.
The researchers created a computationally demanding algorithm to run the auction, to much success. The auction made more spectrum available for faster 5G services, raised $19.8 billion, and netted over $7 billion for the US Treasury. Economist Paul Milgrom, one of the auction’s principal designers, was later co-awarded the Nobel Prize in Economics for “improvements to auction theory and invention of new auction formats,” partly because of this work.
Speeding time to high performance computing research with RONIN and AWS
In 2017, Leyton-Brown and Newman returned to the FCC auction design to assess its strength and weaknesses, in part because of new publicly available data about how TV broadcast and wireless carrier bidders responded to the novel auction design. Leyton-Brown and Newman used AI and machine learning (ML) to create a computer-based simulation of the FCC auction using this new data, and then explored the behavior of bidders in different manufactured situations.
However, this research required significant compute time to process the massive scope of data. For scholars at universities in Canada, who must consistently compete with other scientists for limited public resources, especially for high-performance computing, this posed a challenge.
Instead, the research team turned to RONIN, an AWS Partner. The research team accessed RONIN through UBC’s Advanced Research Computing (ARC), whose role is to provide HPC infrastructure and computational support to researchers across all disciplines. ARC played a pivotal role in recommending infrastructure to Drs. Leyton-Brown and Newman.
RONIN’s software solution is built on AWS to provide a simple interface that helps researchers create and control AWS computing resources, set and monitor budgets, and forecast spend—with minimal knowledge of the cloud. With RONIN, researchers can launch a graphical desktop to a Windows workstation one moment and create an autoscaling cluster more powerful than many on-premises supercomputers the next. For the UBC team, this direct, user-friendly access to virtual clusters made research simpler and faster, helped manage costs, and streamlined laboratory workflows.
Plus, using RONIN, the team gained access to multiple high-end graphics processing units (GPUs), which support parallelized computing tasks like ML and scientific simulations. Instead of working with less efficient in-house central processing units (CPUs) or acquiring on-premises GPUs at significant expense, Newman used GPUs on AWS through RONIN only for the time it took to train, optimize, and run the ML model on the large datasets.
With RONIN, Leyton-Brown and Newman were able to run experiments in a single day that would have taken many weeks using other resources. Using RONIN and AWS, the team used 0.5-1 century of compute time – meaning the same task would take a single CPU 50-100 years to complete – about 5-10 times faster than their other on-premises clusters.
Advancing economics with high performance computation on AWS
While local, institutional, and national compute resources may be enough for isolated studies, the computational simulations needed to answer today’s cutting-edge research questions are getting more complex and more resource heavy.
Work by Leyton-Brown, Newman, and other economists demonstrates that there is a value to studying economic markets computationally. For that, sufficient computing resources are required. Tomorrow’s researchers will have even more extensive computational demands and need new infrastructures to support them. To that end, the virtually unlimited infrastructure of the AWS Cloud, made accessible by AWS Partners like RONIN, can help economists and researchers digitally transform the future of economics.
Get started with AWS for research
With AWS, higher education institutions, research labs, and researchers can quickly analyze massive data pipelines, store petabytes of data, and advance research using transformative technologies like AI, ML, and quantum – all while securely sharing their results with collaborators around the world. AWS also provides researchers with access to open datasets, funding, and training to accelerate the pace of innovation.
Find more case studies, on-demand seminars, trainings, tutorials, and more for your research area at the Research and Technical Computing on AWS main page.
Read more about AWS for research:
- Preventing the next pandemic: How researchers analyze millions of genomic datasets with AWS
- Cloud powers faster, greener, and more collaborative research, according to new IDC report
- How to put a supercomputer in the hands of every scientist
- Building a team knowledge base with Amazon Lightsail
- Emerging trends in cloud for advanced research computing
Subscribe to the AWS Public Sector Blog newsletter to get the latest in AWS tools, solutions, and innovations from the public sector delivered to your inbox, or contact us.
Please take a few minutes to share insights regarding your experience with the AWS Public Sector Blog in this survey, and we’ll use feedback from the survey to create more content aligned with the preferences of our readers.