Q&A: Boosting research with supercomputing
A personal computer can display the week’s weather forecast or open a website to refill a prescription, but only a supercomputer can train the machine learning models and run the extensive simulations needed to predict what climate events may take place decades in the future or to identify the molecular mechanisms underpinning a disease. Researchers in any discipline can tackle similar large-scale scientific questions and more with Penn State’s Roar supercomputer, housed within the Institute for Computational and Data Sciences (ICDS).
“The institute and Roar serve as a major resource to the high-performance computing (HPC) and big data research community within the University,” said Guido Cervone, ICDS director. “ICDS aims to enhance inter- and multi-disciplinary research collaborations using computational and data science methodologies, artificial intelligence (AI) and quantum sciences in this rapidly growing technological era.”
HPC typically refers to a large and extremely powerful supercomputer or cluster of computers working together, or a combination of the two, to store and process large datasets. Roar allows Penn State researchers to run large-scale scientific simulations and build machine learning models and algorithms to perform complex statistical and mathematical analyses that aren’t possible on a single personal computer.
Roar, for example, contains more than 40,000 computational cores and offers researchers nearly 20 petabytes of storage — almost as much as the Library of Congress uses to store its digital content collection. It also includes security and compliance measures necessary to run sensitive data. Now, a new access model allows Penn State researchers to use HPC resources on an as-needed basis for their projects, offering options for full allocations of HPC resources for longer-term commitments or a credit-based option for intermittent workflows. The supercomputer is backed by a technical team of experts and a client support team, consultations with the Research Innovations with Scientists and Engineers team, an expansive software stack and custom solutions and services that allow researchers to meet their specific research needs.
ICDS co-hires like Ed O’Brien, professor of chemistry in the Eberly College of Science, and Romit Maulik, assistant professor in the College of Information Sciences and Technology, spoke about how they use Roar’s HPC services in the Q&A below.
Q: Why is HPC important for your research?
O’Brien: My research tries to make sense of what is going on inside living cells, and to be able to do that, we need HPC resources to model what is happening using rules from chemistry and physics. By using HPC, we can gain new insights into potential disease mechanisms, which could open future opportunities for therapeutic development. We could also gain a better understanding of novel aspects of cellular behavior, which could potentially lead to bioengineering opportunities. Recently, my team published a study that found a potential mechanism explaining why some proteins misfold, potentially leading to disease.
Maulik: My research group, the Interdisciplinary Scientific Computing Laboratory, uses a mix of HPC, data science and applied mathematics to create algorithms for weather forecasting and climate modeling on Earth and beyond; to study the space around merging black holes; and to understand how to mitigate disruptions and damage for safe and efficient operation of nuclear fusion reactors like Tokamaks, for example.
Q: How does Roar enable your work?
Maulik: All my work requires training AI models or running simulations of complex dynamical systems that are not possible on a single workstation. We need HPC systems to conduct our research. My research group uses graphics processing units (GPUs), which are used to speed up tasks and calculations and rendering of graphics and video, for training deep neural networks — AI models designed to mimic how the human brain operates — with applications for various computational physics problems. I’m about to start a two-year project with fellow ICDS co-hire Steven Greybush that seeks to improve atmosphere and ocean weather forecasts integrating AI and satellite data into existing forecasting models. Using ICDS resources could help us build better models for forecasting the weather, which could aid in improving the prediction of extreme events and increase the horizon for accurate forecasts.
O'Brien: Roar resources like GPUs and central processing units (CPUs) — the primary circuitry of a computer that executes instructions — allow my team to run simulations with large, publicly available datasets. The system also allows us to construct AI models and use them to better understand cells, their mechanisms and how different cell states emerge in different organisms, all of which has a broader impact on life in different environments. Data availability is growing exponentially and, in my field, doubling every year. HPC is helpful in storing large amounts of research data. For example, in my research, we often deal with very large datasets on the order of tens to hundreds of terabytes. Using HPC resources, specifically Roar, enables my research lab to take advantage of the explosion of data to gain fundamental insights and answer important scientific questions. The U.S. National Science Foundation National Synthesis Center for Emergence in the Molecular and Cellular Sciences at Penn State, which I direct, uses ICDS Roar resources to examine publicly available, large datasets to explore fundamental questions.
Q: Why should researchers use Roar/HPC?
O’Brien: HPC resources can scale up research on an elevated level. Roar can make solving complex scientific problems possible. Roar is a full-service offering and can help carry out large-scale simulations, data analysis, and the development and deployment of machine learning models that otherwise wouldn’t be possible. Roar enables my research team to be able to complete the tasks needed to gain insights into fundamental science.
Maulik: Researchers may need to use HPC to tackle computational problems that are too large, complex or time-sensitive for standard desktop computers. These problems span many different domains such as physics, engineering, biology, medicine, climate science and, most recently and rapidly growing, AI. Researchers need to be able to quickly read and write or rewrite information and applications to conduct their studies. Compute resources like Roar can help researchers do their work more effectively and on a faster time scale.