The University of Iowa is investing in research data analytics and data-intensive computing through hardware and software enhancements and personnel to support research computing needs.
The investments in research computing address technology needs being articulated by campus, including demand for artificial intelligence (AI), accelerated computing using graphics processing units (GPU), and tools for interactive data analytics and data sharing. Improvements to research computing services are designed to accelerate research productivity, decrease administrative burden, and support faculty recruitment and attainment of grants.
“We are seeing a paradigm shift in research computing today,” says Ben Rogers, senior director of Research Services. “Traditionally research computing has provided technology focused around HPC systems that are designed for command-line usage and batch computing. As data sets are growing in size and new methods such as machine learning become mainstream across disciplines, we are seeing greater need for more user-friendly graphical and interactive systems. Investments and changes underway at the UI will ensure we are successful in supporting this paradigm shift.”
Supporting growth in data science
UI is taking a multi-faceted approach to meet expanding needs in data science.
Research Services is expanding research computing support to include popular data-science platforms such as R and Python. Hiring is currently underway for staff to support this effort and currently an early fall 2019 launch is planned.
Work has already been done on the HPC systems to improve access and performance of data-science tools. New toolsets related to deep learning, such as TensorFlow, have been installed and optimized. More work on software support for complex software stacks, such as AI environments, starts this spring.
To improve campus data-sharing options, particularly for large data sets, the Globus software suite is being deployed and is currently scheduled for launch by summer 2019.
Enhancements in HPC and GPU
UI high-performance computing (HPC) resources now serve about 900 users in 100 departments.
Among the users are professors Bryce Dietrich, who uses HPC to probe the predictive power of motion and vocal pitch in politics and court proceedings, Bill Barnhart, who uses HPC to gauge earthquake risks, and Mike Schnieders, who uses HPC to categorize deafness by a wide range of genetic traits and environmental factors so precautions can be taken to protect hearing.
In 2017, the UI commissioned Argon, the third iteration of its high-performance computing (HPC) resource. Soon, changes to the HPC environment will reduce the number of HPC environments from two to one, eliminating confusion and administrative overhead by providing a single location for HPC users.
New GPU capabilities added in August 2018 allow the use of lower-cost, higher performing graphics processing hardware, accelerating work in areas such as deep learning and molecular modeling. For some workloads, the new GPUs can be 20 times faster than traditional central processing units (CPUs).
“The UI HPC system allows the datasets to be processed in parallel, reducing a multi-week processing task into a single day,” says Joseph Reinhardt, Iowa Informatics Initiative affiliate and image analysis group leader of the Iowa Institute for Biomedical Imaging. “Furthermore, the graphics processing units available on the HPC system have allowed us to train state-of-the-art convolutional neural networks in less than a day. This would not be possible without the high-memory GPU cards on the HPC system.”
That sentiment is echoed by Milan Sonka, professor of electrical and computer engineering, director of the Iowa Institute for Biomedical Imaging, and College of Engineering associate dean for graduate programs and research.
“After decades of continuously struggling with inadequate computational power which adversely affected the global medical image analysis research community, such revolutionary methods are finally within reach,” Sonka says, noting that there is little doubt among computational medical imaging specialists that the trend toward the use of accelerators will continue.
“Maintaining a globally-competitive computational portfolio will not be easy or inexpensive for UI or Information Technology Services, but it is crucial if we hope to maintain a performance edge.”
OneIT supports consumer GPU cards, lowering the cost per GPU by up to 80 percent over traditional enterprise cards. In an initial purchase, savings from using consumer cards can exceed $300,000.
Research support and efficiency
Beyond computational and data-science platforms, OneIT is working to accelerate research through efficiency and reducing burden on researchers. Current areas of focus include self-service tools, systems for managing research funding, and technology compliance support.
OneIT has launched a number of self-service tools but one of particular value for the research community is the ability for faculty and staff to sponsor guest HawkIDs in real time. It provides a quick, easy way to create collaborator accounts for many services offered by the UI.
The next phase is underway to provide self-service access management via the web. One target for this ability is the Research Data Storage Service. Used by more than 500 research groups, the service is available to all faculty and provides network file shares with backups, replication, recovery, and auditing.
To reduce the burden of paperwork for researchers and staff, Research Information Systems teamed up with campus partners to improve processes and develop tools to assist in research administration.
The eDSP (Division of Sponsored Programs) application streamlined the review and tracking of proposals and awards for research funding. It enabled a paperless process and remote work capabilities for the Division of Sponsored Programs while increasing transparency for the research community. On average, 550 proposals and awards are reviewed through eDSP each week.
A sub-award application builds upon this work, facilitating the process of submitting, reviewing, and tracking outgoing sub-award requests for external research collaborations. It utilizes the same work queue to accurately track sub-awards from submission to approval, and integrates with the purchasing system, currently managing over $37 million in outgoing funds for cross-institutional collaboration.
OneIT is also is also working to help researchers through the increasing complexity of technology compliance. A research technology compliance specialist was hired to identify IT solutions that work for the community while adhering to regulations.
The specialist’s work has focused on process improvements in Institutional Review Board and security plans, helping researchers through the compliance process, and certifying services for sensitive data.