SciDAC National Fusion Collaboratory Project Sees Gains With Globus Toolkit

Magnetic fusion energy research is getting a boost from deployment of the Globus Toolkit TM. Project leaders hope the popular toolkit's Grid software and services will let them collaborate better -- and in different ways -- with partners across the U.S. and overseas. The National Fusion Collaboratory (http://www.fusiongrid.org/) is funded by the U.S. Department of Energy's SciDAC program to create and deploy collaborative software tools for magnetic fusion research. The FusionGrid project seeks to promote more efficient use of experimental facilities and more effective integration of experiment, theory, and modeling. David Schissel is a FusionGrid principal investigator from General Atomics, where he manages the Data Analysis Applications group, part of the Theory and Advanced Computing division. "We asked ourselves, 'How can a distributed community like ours make collaboration more fruitful?'", said Schissel. Through contact with Rick Stevens, he became aware of new types of tools being created in the Argonne National Laboratory (ANL) Mathematics and Computer Science (MCS) division, which Stevens directs. MCS is home to the Distributed Systems Laboratory (DSL) -- a lead partner in the Globus Project (http://www.globus.org) -- and the Futures Laboratory (FL), which develops advanced visualization technologies that are also proving useful to fusion researchers. The discussions with ANL led to a successful proposal for three years of SciDAC funding that began in late 2001. The Globus Project's Ian Foster and Kate Keahey are on the FusionGrid team for the DSL, as are Stevens and Mike Papka for the FL, along with others from MIT, the Princeton Plasma Physics Laboratory, Lawrence Berkeley National Laboratory, Princeton University, and the University of Utah. With principal support from the DOE Office of Fusion Energy Sciences, the U.S. magnetic fusion community relies on three tokamaks, at MIT in Boston, Princeton in New Jersey, and General Atomics in San Diego. A tokamak is a large doughnut-shaped device in which magnets are used to confine extremely high-temperature plasmas. The ultimate goal is to create an economical, sustainable, environmentally friendly energy source. "Our users are geographically distributed across the country," he said, "but experimental data are centralized at those three sites. To begin, we're using the Globus Toolkit for remote computation -- including secure data access -- but our eventual goal is to be able to perform the majority of experimental data analysis remotely." Each plasma exists for only about 10 seconds, with 15 to 20 minutes between plasmas, which presents a very intense data-analysis period. Grid computing can deliver more-detailed and immediate feedback about the plasma behavior, letting fusion scientists make adjustments on the fly, yielding much more-productive experiments. "What if you did a day of 30 plasmas but couldn't analyze resulting data until the following week?", said Schissel. "That's not an efficient use of experimental time. The Grid way has the potential to be a much better use of time in terms of equipment and people." Using the Globus Toolkit combined with visualization tools from the Futures Lab and other project members, fusion scientists at their home institutions can retrieve and analyze remote data during experiments in progress, then give tokamak operators feedback in time to shape the next plasma run. The 44th Annual meeting of the Division of Plasma Physics during the week before SC02 brings together over 1,000 fusion energy scientists. The meeting will feature two invited presentations of results from FusionGrid's remote computing capability and remote visualization tools, which will also be demonstrated there. "We're excited that, within one year of start-up, FusionGrid is producing results useful to the fusion community," said Kate Keahey of Argonne National Laboratory, a Globus Project member of the FusionGrid team. "Not every site can afford its own cluster or the manpower to maintain one, so the good news is that if you have Grids, you don't have to." MDSPlus (www.mdsplus.org) is the most widely used software for managing magnetic-fusion experimental data. It is currently installed at over 30 sites on four continents. The FusionGrid project has "Globusized" the MDSPlus tools, which were developed ten years ago by MIT, the Center for Nuclear Research in Padua, Italy, and Los Alamos National Laboratory. FusionGrid principals hope to bring experimentalists and theorists closer together by using the Grid-enabled MDSPlus as a common platform for organized data storage, something the theoretical community has lacked previously. "If I'm working at Princeton and have a computer code there, before I would have brought the code home and it would then have to be maintained at two sites. With Grids, we can share code across 20 to 30 sites," said Schissel. "I might not have access to a large Beowulf cluster at my home institution. With the Globus Toolkit's authentication and authorization services, we're making progress. Another program, TRANSP from Princeton, has been used since the mid 1980s for data analysis and simulation. TRANSP runs on Linux clusters, and a Globus Toolkit-enhanced version is now being demonstrated at scientific meetings and is producing scientific results that have been presented at the 19th International Atomic Energy Agency Fusion Energy Conference, in addition to the national plasma physics meeting demos and presentations, according to Schissel. Now they're working on NIMROD, a tool that generates 3D macroscopic models of fusion plasmas. "Our funders ask us to identify positive results that FusionGrid research is having for scientists," he said. "The Grid versions of TRANSP and MDSPlus are good examples. So far the overhead is a bit high, though, because the ratio of programmers to scientists is too high. We anticipate that -- as Grid technology matures and fusion scientists gain experience with it -- in time, this ratio will decrease and the entire field will see great benefits." Schissel will describe FusionGrid during presentations at the upcoming SC02 conference in Baltimore, November 16-22. He will appear at booths hosted by SciDAC, ANL, the University of Utah, and the University of California at Berkeley.