September 09, 2005
Thomas Zacharia, Oak Ridge National Laboratory's associate laboratory director for Computing and Computational Sciences, is leading ORNL's multi-year initiative, on behalf of the Department of Energy, to develop the world's most powerful computing resource for open science. In this exclusive interview with HPCwire, Zacharia provides updates on the lab's Leadership Computing Facility and strong involvement in the field of climate science.
HPCwire: You're giving a keynote speech at the Computing in Atmospheric Sciences workshop in Annecy, France next week. What will you talk about?
Zacharia: I'll be talking about the DOE's Leadership Computing Facility at Oak Ridge, especially our high-level goals and expectations for contributing to the field of climate science, and about leadership computing in general as a necessary enabling technology for advances in climate science. I'll also talk about ORNL's contributions to this field. For quite some time, we've had people doing breakthrough work on the computational side and the experimental side, including things like the economic impact of global warming and other climate change. We have a long-standing relationship with Warren Washington and his group in NCAR's Climate Change Research Section, and ORNL has made significant contributions to the runs for the IPCC [International Panel on Climate Change] Fourth Assessment on global warming that is due out in 2007. About one-third of the simulation work for the international initiative has been done at ORNL. We're involved in a lot of collaborative activities in climate science, in the U.S. and around the world.
HPCwire: How did Oak Ridge and other DOE labs get involved in climate modeling? Where does that intersect with energy?
Zacharia: DOE has a long history of involvement in important climate science issues, such as evaluating the carbon management problem. Most scientists now agree that current sources of mass energy production and usage have a significant impact on the environment and on climate. We have recently initiated a line of research that takes predictions of global warming, in terms of heating and cooling days, and predicts energy usage and energy technology development. Oak Ridge has had a strong institutional emphasis on climate science for a long time.
HPCwire: Which areas of climate modeling and other atmospheric science is Oak Ridge involved in?
Zacharia: As I mentioned earlier, we did about one-third of the runs for the most recent IPCC simulation of global warming. ORNL has a very strong Computational Climate Dynamics group that is doing breakthrough work in coupled climate-carbon simulations. John Drake, David Erickson and others have looked closely at this. John directs the Computational Climate Dynamics group and David heads our Climate and Carbon Research Institute, which uses computational methods to evaluate how the global climate and carbon systems interact with natural and anthropogenic processes. They're trying to get a better understanding of how the Earth's climate responds to physical, chemical, and biological changes produced by global alterations of the atmosphere, the ocean and the land.
HPCwire: How does HPC support ORNL's climate work?
Zacharia: Climate science was one of the earliest and most insatiable consumers of high-performance computing resources. HPC today is a key component in simulating climate change and in performing diagnostics based on climate simulations. Beyond computing resources, there's also a tremendous, growing need for storage of historical, contemporary and forward-looking climate data. ORNL has vast amounts of atmospheric and other climate data in storage.
HPCwire: You mentioned Oak Ridge's Climate and Carbon Research Institute is heavily involved in research related to global warming. Can you say more about that?
Zacharia: ORNL is a $1 billion-per-year research organization, and climate research is an important component of our research activities. Our Climate and Carbon Research Institute takes multiple approaches, including not just computing but also taking atmospheric radiation measurements and doing a lot of work on the economics of climate and carbon management. CCRI was formed as an integrating principle for people approaching this problem from various perspectives and disciplines. It's akin to people touching different parts of the elephant and then communicating with each other to assemble the big picture.
A lot of interdisciplinary collaboration is needed for good progress in climate science. Because this collaboration at CCRI includes economic projections and forecasts, for example, we've made better progress in predicting energy changes and demand over a 100-year cycle. Climate prediction is a classic example of an experiment you can't run in nature. You need to simulate it. It helps a lot to have many decades of hard experimental data in storage, as we do at ORNL. This is important because it allows you to run the simulation backwards in time for validation before you project to the future. These huge simulations require a lot of computational power. Our NCCS Leadership Computing Facility is an extremely powerful tool to place at the disposal of some of the world's most creative scientific minds in the field of climate science.
HPCwire: The International Panel on Climate Change's so-called Fourth Assessment is due out in 2007. I gather the main purpose is to take a more detailed look at global warming. Can you give us an update?
Zacharia: Sure. The IPCC runs were completed earlier this year. A significant portion of the runs, which in total simulated 1,100 years, were calculated at ORNL's Leadership Computing Facility using the CCSM2 climate model in a high-resolution atmospheric configuration. A couple of hundred papers have already been published on the IPCC runs, which look at climate as a function of atmospheric carbon dioxide and other greenhouse gases. Among other things, the runs confirmed that smaller concentrations of these gases result in less warming.
HPCwire: Oak Ridge is heading toward petaflop computing capability by 2010. What will that make possible in climate science that can't be done on today's best HPC systems?
Zacharia: We plan to have petascale capability sometime in the 2008-2010 timeframe. Our experience, not only in climate science but other fields, is that as computing ability increases, two things improve. First, you can go to more accurate models with finer grid resolution. In climate science, people are talking about moving from today's 100-kilometer global models to 30-kilometers and eventually five-kilometer models. Second, you can add many variables that researchers don't include today because they don't have enough compute power. There are about 100 non-linear variables we'd like to see included in climate simulations. In particular, we envision a strong need to include a fully active carbon cycle inside the more traditional physically-based climate simulations. In climate science today, as in many other compute-intensive disciplines, we have to practice deliberate reductionism in order to complete simulations in reasonable periods of time.
HPCwire: You're also working on how to create and control plasma energy, which is the same energy the sun produces through an atomic fusion reaction. What can you tell us about this?
Zacharia: ORNL is one of the leading participants in the multibillion-dollar ITER project, which is a major collaboration involving the U.S. and various nations in Europe and Asia. ITER stands for International Thermonuclear Experimental Reactor. Starting in 2006, the goal is to develop, by 2016, a working plasma energy reactor based on the same atomic fusion process that occurs in the sun. The ultimate goal is for this to lead to fusion power plants to produce electricity. This form of energy is almost limitless and it's very clean. Using our new Cray X1E machine, ORNL has been able to do the fastest and most-detailed simulations to date of the waves that will be used to control the plasma in the ITER reactor.
The U.S. is making a significant contribution to the ITER project, and the Princeton Plasma Physics Laboratory and ORNL are leading the U.S. effort. PPPL and ORNL won the competition to manage the U.S. ITER project office. In addition, ORNL was just awarded the contract to carry out the Fusion Simulation Project that will evaluate the impact of the ITER reactor.
HPCwire: How are things going with the installation and ramp-up of your big Cray systems?
Zacharia: Things are going very well. The Cray X1E has been terrific. Our experience with the predecessor Cray X1 computer was extremely good, so we decided to upgrade to an 18.5-teraflop X1E. We subjected the X1E to rigorous acceptance tests, and it passed with flying colors. We already have five grand challenge projects running on it and are seeing a lot of breakthrough science that has not been possible on other contemporary HPC systems.
Our experience on the Cray XT3 has been really good so far. Cray was able to deliver it on time and we have been putting it through its paces. We also have very rigorous acceptance tests for the XT3, and we expect to accept the system soon. It's different from the situation with the X1E. We knew a lot about the X1E because of our experience with the X1. We didn't have prior experience with the XT3. We have 56 cabinets of the new XT3 installed and running, and it's already delivering new science for us. We're very excited about this system. As we get more experience with it, it will continue to improve as a production resource for our Leadership Computing Facility.
HPCwire: Will the U.S. and international climate communities have any access to ORNL's computing resources?
Zacharia: I'll be surprised and disappointed if we don't get proposals from the broader climate science community to use our new HPC resources. ORNL has provided substantial support to the climate community in recent years and I expect this to continue. DOE has set aside 10 percent of our Leadership Computing Facility resources for the INCITE program, which seeks computationally intensive research projects of large scale that can make high-impact scientific advances through the use of a large allocation of computer time and data storage. The INCITE program is open to proposals from anyone and everyone, including industry.
HPCwire: Is there anything else to add?
Zacharia: Just that we're really excited about the prospect of taking computational science to substantially new levels. It's a great opportunity for science. We are on the cusp of an era where tremendous new discoveries will be realized using computing. We in the HPC community are facing some real challenges in terms of effective software, effective hardware and effective facilities at this scale. My sense is that at the petascale level and beyond, these challenges will have to be addressed much more interdependently. It is a great opportunity and a responsibility to do it right.
10/30/2013 | Cray, DDN, Mellanox, NetApp, ScaleMP, Supermicro, Xyratex | Creating data is easy… the challenge is getting it to the right place to make use of it. This paper discusses fresh solutions that can directly increase I/O efficiency, and the applications of these solutions to current, and new technology infrastructures.
10/01/2013 | IBM | A new trend is developing in the HPC space that is also affecting enterprise computing productivity with the arrival of “ultra-dense” hyper-scale servers.
Ken Claffey, SVP and General Manager at Xyratex, presents ClusterStor at the Vendor Showdown at ISC13 in Leipzig, Germany.
Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?