The Year in Perspective

One of the scientific projects at NERSC aims to increase understanding of the complex chemical processes that occur during photosynthesis, taking in carbon dioxide and producing oxygen in the process. This project, led by William Lester of UC Berkeley and described in detail in this annual report, is important on several levels. First, plants and bacteria are the world’s foremost means of “carbon sequestration,” or removing carbon from the atmosphere in the form of CO₂ — a process which has enormous implications for climate change and global warming. Additionally, photosynthesis is an example of fundamental electron chemistry and is an efficient energy transfer system — processes which are important in many areas of scientific research. But most importantly for the NERSC community, this project was selected under a new competitive program, entitled Innovative and Novel Computational Impact on Theory and Experiment (INCITE), conceived by Dr. Raymond Orbach, Director of the DOE Office of Science. The goal of the program was to select a small number of computationally intensive large-scale research projects that can make high-impact scientific advances through the use of a substantial allocation of computer time and data storage at the NERSC Center.

The first full year of the INCITE program at NERSC in 2004 brought some truly outstanding scientific results using our computational resources. In addition to the better understanding of energy transfer mechanisms in photosynthesis, INCITE researchers investigated stellar explosions and the contribution they make to the abundance of intermediate-mass elements. They also used NERSC to gain greater insight into fluid turbulence and mixing at high Reynolds numbers. From accomplishments such as these, it is clear that we really have entered the new era when computational science is an equal partner with theory and experiment in making scientific progress. All the INCITE results and dozens more are described in our annual report. They demonstrate that NERSC continues to provide one of the most effective and productive high-end capabilities for computational science worldwide.

Supercomputing continues to evolve at a rapid rate. While there are several systems at other sites that claim records of one sort or another, our 10 Tflop/s Seaborg system continues to break new ground when it comes to reliability and productivity using a unique resource. In 2004 NERSC has moved aggressively to devote a greater share of its processing time to jobs running 512 processors or more. On average, now more than half of all jobs on Seaborg fall into this category, and towards the end of the year these large jobs used around 78% of the system, a dramatic improvement over last year. One of these jobs was a calculation of an entire year’s worth of simulated data from the Planck satellite, which ran on 6,000 processors in just two hours.

NERSC users already have the benefit of exploring the scalability of their applications to thousands of processors, and consequently will be ready for the next generation of supercomputer platforms with tens of thousands of processors. Scaling to larger numbers of processors is not only a challenge for algorithms and applications, but it also exposes other limits implicitly built into our thinking and our systems. For example, when trying to tune the performance of codes running on large numbers of processors, most of the existing performance tools exhibited large overhead, and NERSC staff had to develop Integrated Performance Monitoring (IPM). IPM has extremely low overhead and is scalable to thousands of processors. Of course IPM was applied to analyze the INCITE applications, and helped to make the most effective use of their allocations. This is a beautiful example of how the integration of powerful computing systems and excellent staff at NERSC are creating productivity gains for computational scientists that are not easily obtained elsewhere.

On the national scene, a number of new activities that were initiated in 2002 and 2003 were completed in 2004 and will hopefully make a profound impact on computational science and high performance computing in the U.S. The High End Computing Revitalization Task Force (HECRTF) report was released in March 2004, providing the blueprint for further development of high-end computing in the U.S. and defining areas of collaboration for Federal agencies. The second volume of the “Science-Based Case for Large-Scale Simulation” (SCaLeS) report demonstrated the wealth of applications and the potential of scientific progress enabled by next-generation platforms. Finally, the National Research Council study “Getting Up to Speed: The Future of Supercomputing” was released in November 2004, and makes a powerful case for the importance of supercomputing both for conducting basic scientific research, as well as for ensuring the economic and physical well-being of the country. These reports give us a glimpse of what simulations are possible at sustained speeds in the range of tens to hundreds of teraflop/s, and lay out the critical research issues that the community needs to address in order to reach petascale computing performance. This is exciting news, because we are thinking big again in supercomputing. It remains to be seen how well this enthusiasm will translate into future budgets, but it is highly satisfying to see so many NERSC users being deeply engaged in these strategic processes and helping to build our future. Clearly the DOE computational science community, in particular NERSC users, are ready to take the lead in moving up to the next level of computational science.

Several steps towards this bright future will happen at NERSC in 2005. NERSC will acquire and put into production two new computing systems that will take over some of the existing capacity user workload from Seaborg, making even more time available on Seaborg for large-scale capability computing. This will be just an intermediate step towards our next big acquisition, NERSC-5, which will also be initiated in 2005. Thanks to the support from our program management at the DOE Office of Advanced Scientific Computing Research, the NERSC budget has been set at a level that makes all these ambitious plans feasible. Thus we are confidently looking forward to another year of both scientific and computing accomplishments at NERSC. As always, this progress would not be possible without the NERSC staff, who continue to tirelessly dedicate their time, skill, and effort to make NERSC the best scientific computing resource in the world. Our special thanks to all of you.

Horst D. Simon
NERSC Center Division Director

William T. C. Kramer
Division Deputy and NERSC Center General Manager