High-performance computing (HPC) for research is notorious for having steep barriers to entry. For this reason, high-tech disciplines were early adopters, have used the most cycles and typically drove hardware and software purchasing decisions. But as more fields engage with data-intensive research and artificial intelligence (AI) workflows, data-center hardware and software landscapes are changing to meet a larger and more diverse community of practice.
Iowa's new Interactive Data Analytics Service (IDAS) is an HPC resource that supports large-scale and collaborative data analytics workflows involving RStudio for R and Jupyter Notebook for Python (but not limited to Python). Applications via the IDAS interface look and feel like they do on a regular workstation, but with access to thousands of times the computing power. IDAS has its own HPC and graphics processing units (GPUs) and allows users to perform interactive data analysis tasks with applications used for machine learning and AI. In the future, the UI Research Services development team will assist with custom environments. There are plans to add a feature that allows researchers to access Iowa’s Argon supercomputer if more power is needed, or to purchase commercial cloud if locally-hosted HPC isn’t enough.
While RStudio for classroom use is available, RStudio commercial licensing terms changed and pricing increased midstream as IDAS was being developed. Therefore, until a more affordable and dynamic license distribution method is available, IDAS employs the free RStudio classroom license. In the future, a remote desktop feature will be added.
IDAS especially appeals to those who may not have used HPC in the past, but whose data-intensive research would benefit from advanced computational capabilities. Such disciplines are often referred to as, “the long tail of science,” and many have begun to engage with AI methodologies which make their work more computationally-intensive than ever. While they aren’t always the biggest compute consumers, their collective needs represent critical mass that IDAS accommodates quite well.
Experimentation within IDAS is safe with associated storage options that are appropriate for most classifications of data. Sensitive data should be evaluated for IDAS consideration on a case-by-case basis. “If you aren’t sure about your data’s classification, you can arrange a consultation with our compliance specialist,” said Research Services Director Danny Tang.
IDAS Jupyter Notebook serves both general research and classroom needs which were optimized during its pilot phase. A dozen faculty and staff from six departments test-drove the research platform for several months, and classroom attributes were trialed by students in Data Scientist Kang Lee’s (UI Research Services; College of Business) summer business analytics class.
At the time of the mid-August launch, eight classroom instances were registered, including one by Professor Paul Gowder (UI College of Law, photo on right). Gowder’s research involves the impact of machine learning and predictive modeling on both the practice and concept of law. “I’m planning to use IDAS in my ‘Introduction to Quantitative and Computational Legal Reasoning’ course, and may also utilize it for my research involving AI to understand contracts and records from publicly-available court documents,” said Gowder.
Teaching Assistant James Kent (UI Interdisciplinary Graduate Program in Neuroscience) aides Professor Michelle Voss with Iowa’s structural/functional Magnetic Resonance Imaging (MRI) methods course (photo taken during their hackathon on left). “Our two biggest challenges were installing software on heterogeneous hardware, and random-access memory (RAM) restrictions when running a virtual machine,” said Kent. “With Jupyter Notebook on IDAS, the HPC is handled in the background; this provides students with enough computational power to complete full analyses on real MRI data, not the toy datasets we always curated for this class,” he added.
Kent explained how useful he imagines IDAS will be for workshops, too. Since most are one to three hours in length, less instructional time will be spent installing software and/or learning how to master the hardware. “Some students get discouraged if the preparation phase is too tedious; it presents an unnecessary hurdle,” he said. “If students feel this time is wasted, some may skip the workshop, leave early or walk away with the impression that programming is inaccessible to them,” he added.
Outside of the formal classroom setting, Kent thinks IDAS will be perfect for hackathons. “They’re often about prototyping and Jupyter Notebooks provides a great platform on which to interact with code and receive immediate feedback,” said Kent. In the past, a downside of hackathons was that machine learning models can take a lot of RAM and/or use significant GPU power. HPC is the answer for increased RAM and GPU usage, but mastery of Jupyter Notebooks was difficult with the traditional HPC interface. Kent expressed appreciation for IDAS’ development, and sustained back-end management when he said, “IDAS combines Jupyter Notebooks’ ease of prototyping with the power of HPC that runs and scales ‘automagically’ in the background!”
Brandon LeBeau (UI College of Education, photo on right) uses IDAS in an introductory statistics course that allows future educators to explore statistical concepts, remove assumption barriers in the statistical analysis, and introduces them to best practices surrounding script-based analyses. “The goal for this course is for students to learn how to reason with data and make appropriate statistical/data arguments, vs. learning how to program,” he said, and added, “This hands-on approach will help them on their journey to becoming stronger data connoisseurs.”
The computation component in IDAS’ Jupyter Notebooks instance, with an R kernel, gives students access to a modern computing environment without requiring an instructor who doubles as an information technology (IT) professional. “The user experience is also oblivious to compute functions managed by Research Services on the back-end; otherwise the need for programming skills would become a barrier to entry for many students and faculty alike,” said LeBeau.
IDAS was the brainchild of former UI Research Services Director Ben Rogers who aligned the necessary collaborations and allocated funds to make his vision a reality. It was developed by Lee and Interactive Systems Lead Cody B. Johnson. Research Services would like to thank Chris Wilkins and Kevin Zhu of the UI Core IT Facilities Group for their help with deploying IDAS.
“While ten other universities that we know of have similar environments, Iowa is one of the first to host a large-scale, cloud-enabled and expanded HPC resource,” said Lee. “We’re excited about the number of Iowa colleges and departments that engaged in our pilot, and some of the novel applications they’re pursuing on IDAS,” he added.