University of Tennessee (UT) chemistry professor Robert Harrison was recently named director of the Joint Institute for Computational Sciences (JICS), a partnership between UT and Oak Ridge National Laboratory (ORNL) that seeks to further all elements of computational science research. Below he sheds some light on himself, his organization, its mission, its future, and its challenges.
SC Online: Tell us a little about yourself and your research interests.
Dr. Harrison: I’m a theoretical and computational chemist, with a long-standing interest in high-performance computing. To translate even the most elementary theories into useful tools for practical chemical discovery, you have to do large-scale computation. Chemistry itself is a very mature discipline, and any new advance in computational abilities is really layered on top of a long history and foundation of other methods, so computational chemistry packages are huge. NWChem, one of primary research codes, is pushing five million lines of code now, and that represents hundreds of man-years of effort, so even though our primary interests are theory and chemistry, as a community computational chemists are forced to be good software engineers and computer scientists as well.
In all of these advances applied math is pivotal, both in terms of manipulating the equations and in terms of efficient and robust realization of those equations. A great deal of my recent career has been mining the interfaces between chemistry and computer science and chemistry and applied math. My research activities are split between ORNL and the university. Presently at UT, I have grad students investigating core-hole X-ray spectroscopy, developing numerical solvation models, exploring heavy element chemistry, looking at the behavior of molecules in intense electric fields, and developing tools for analysis. All of this is done in collaboration with experimental or other theoretical chemists/physicists who are more applied in their interests because in any modern science you can no longer just be a pure theoretician or a pure computational person—you have to be relevant to the scientific domain and that involves speaking the language of experimentalists and actually solving real physical problems. You can summarize my philosophy by what I always say to an incoming young staff person: “Do good science and have fun. The two go together.”
SC Online: Can you elaborate on your organization and its unique structure?
Dr. Harrison: JICS is the second-oldest joint institute between UT and ORNL and is housed in a nice, university-owned and managed building on the ORNL campus. This year we are exactly 20 years old, a cause for celebration in itself. JICS is on a very strong upward tick right now. We are proud to be the home of the National Institute for Computational Sciences (NICS), the National Science Foundation’s (NSF’s) largest supercomputing facility which is run by UT. NICS delivers 65 percent of the NSF’s computing cycles, distinguished by the fact that the organization’s premier system, Kraken, is the NSF’s most scalable supercomputer. A large fraction of Kraken’s delivered time is for applications using a substantial fraction of the entire computer, in sharp contrast to the other centers for which the peak is around 256 nodes. Around this we’ve nucleated many other projects and they are all continuing to grow. Perhaps the latest, most substantial development is the recent award to serve as the operations center for the NSF’s Extreme Science and Engineering Discovery Environment (XSEDE) project, which has now replaced the TeraGrid program. JICS is also home to the Remote Data Analysis and Visualization (RDAV) facility, of which Sean Ahern is the Principal Investigator, and multiple other research programs.
Through JICS we have three faculty lines in addition to my joint appointment with the department of chemistry, and those three lines are split into six joint appointments. We will be having a cluster hire for the three remaining empty joint appointments in the near future. There are also two other joint appointments that aren’t funded through this mechanism. These are people in computational biology, visualization, mathematics, materials science, chemistry, and computer science, all of which leads to a very fluid research environment. If we look at the relationship between ORNL and UT, although these are very different organizations with different missions, they have a shared objective to advance the quality of their research and the breadth and depth, in particular, of their computational science.
JICS is by far the largest of UT and ORNL’s joint institutes and is distinguished by the fact that it has this large national facility, NICS, embedded within it. A shared vision by me, UT, and ORNL is to have JICS become an incubator for many research activities and collaborations that are in a very broad sense somehow associated with computation. We’re not just about computation at the high end, but clearly that’s a focus. Overall, the computational aspect goes from students using laptops in classes all the way up through people migrating onto clusters to data-intensive projects, which may not necessarily involve a lot of computing but never the less represent challenges in translating ideas into implementation. At JICS we have resources to support startup activities, grad students, and even faculty time as they begin new ideas. In addition to being home to these joint appointments, we are in the process of trying to create a more fluid environment, one in which a steady stream of researchers rotates through JICS along with their grad students, creating a multidisciplinary environment where students can come and interact with their peers from other disciplines. Unfortunately, there are significant barriers to achieving this, such as the ORNL campus being more than 20 miles from UT. To address this JICS now has space for around 15 people in the Claxton building. This represents the first time we’ve had a substantial presence on the UT campus. The offices are conveniently co-located with Jack Dongarra’s Innovative Computing Laboratory and Lou Gross’s National Institute for Mathematical and Biological Synthesis project, setting the stage for Claxton becoming the center of computational science on campus.
SC Online: How is JICS involved in educating the next generation of computational scientists?
Dr. Harrison: UT is actively developing a graduate program in computational science, viewing the field in the way it cuts across all scientific domains as a critical research element as UT looks to increase its standing in the national ranking. And JICS is central to this because it’s one of the mechanisms with which the university can access ORNL’s very deep capabilities in computational science. Another critical component is the Center for Interdisciplinary Research and Graduate Education (CIRE), the new graduate degree program established by UT and ORNL and headed by Lee Reidinger.
All of UT and ORNL’s Joint Institutes receive money from the university, sufficient to fund the operation of the center and two research scientists. Christian Halloy, a computational physicist and a former director of JICS, and Kwai Wong, who specializes in mechanical engineering, are both excellent and very experienced researchers. They are actively involved in engagement with campus, offering training and working with graduate students in specific science areas and bringing researchers closer to NICS, touching all aspects of computational science life on campus. This includes curriculum development and design of a graduate program in computational science, as well as the existing interdisciplinary graduate minor in computational sciences. JICS used to be an informal entity based on campus but it briefly disappeared from campus in the period that ORNL was establishing its leadership computing facility (OLCF), and subsequently, UT and ORNL established NICS. Now we’re back on campus and stronger than ever.
The question is: What do we want to ultimately accomplish? As I stated earlier, up until recently, ORNL and UT largely embraced different missions computationally. Several things changed that. One was the success, primarily by Thomas Zacharia, in landing the award from NSF that created NICS. The second critical component is the realization on campus of the importance of computational science and the shared understanding of ORNL’s desire for a strong university partner in computational science. One key element of that is our focus right now on research and graduate training and education. If you look at the students coming into our graduate programs from the undergraduate world, those that haven’t already had some exposure to computation, such as thinking algorithmically, solving problems on the computer, and the little bits of applied math that you need to understand all of that, those students have lost a year or two of productivity at the graduate level. But it’s not only the undergraduate students coming into graduate school that have this issue, it’s also our undergrads going off into the larger world. Industry and many other aspects of the commercial world use simulation and computation in diverse ways. So I think a very important role for JICS is to ensure that these concepts go beyond the graduate curriculum and seep into the various undergraduate programs, helping the students that leave UT to be nationally competitive.
SC Online: What is JICS’s main goal for the future?
Dr. Harrison: Five years from now I would like to see a stronger faculty research component. It’s very credible that as we succeed in growing external research funding that the university will reciprocate with more faculty lines. The research funding both at UT and ORNL already supports numerous research faculty, so the number of people supported on soft money will continue to grow. Now that we have a physical presence on campus and there is room for growth there, we can create this environment of multidisciplinary research, and that I think will be a source of huge innovation with unexpected outcomes. Clearly the major goal is to continue to expand the NICS supercomputing center and to that end we have technology partnerships with multiple companies in the computer industry. From the UT perspective a major goal on the horizon is that once computational science has matured to the point where it can stand alone, the university will need a new building and industrial partners that are substantially sponsoring collaboration with UT. This is an objective that is shared with senior UT leadership, not just my personal wish list.
SC Online: What’s been your biggest challenge so far?
Dr. Harrison: The biggest challenge, by far, has been the sad passing of former NICS Director Phil Andrews http://www.utk.edu/tntoday/2011/02/24/mourning-passing-nics-director/. His leadership was an enormous loss to the organization and the supercomputing community in general. However, Patricia Kovatch’s leadership in the interim proved pivotal to JICS’s continued success. She has now moved on and we’ve backfilled management, and we’re working with our project partners at XSEDE to move things forward. That’s a huge opportunity and responsibility for us and we’re deeply committed to the success of that project.
SC Online: What are JICS’s greatest strengths?
Dr. Harrison: On the research side we have a very strong faculty component backed up by the immense experience and abilities of Christian and Kwai, as well as some very strong research scientists supported by both NSF and other agency projects based at UT, and DOE projects primarily at ORNL. Regarding NICS, then, without a doubt, our greatest strengths are operations and user support. NICS contributes exceptionally to both the operation of Kraken and the XSEDE project. In addition to standing up Kraken on-time and on-budget, no small feat in the supercomputing community, NICS delivers fantastic user support that users have time and time again expressed immense appreciation for. Finally, JICS, as one of the five joint institutes, brings together the resources, capabilities, and skills of both UT and ORNL, providing leadership in both organization and creating opportunities for collaboration and growth on both sides.
SC Online wishes to thank Robert Harrison for his time and insight. It would also like to thank Scott Jones for his assistance.