Validating DOE's Office of Science "capability" computing needs [report]

Peter L. Mattern, William J. Camp, Robert W. Leland, Edwin Howard Barsis
2004 unpublished
6/28/04 7 II. Abstract A study was undertaken to validate the "capability" computing needs of DOE's Office of Science. More than seventy members of the community provided information about algorithmic scaling laws, so that the impact of having access to Petascale capability computers could be assessed. We have concluded that the Office of Science community has described credible needs for Petascale capability computing. 6/28/04 19 Applications Summary -Examples • Combustion: Petascale computing
more » ... Petascale computing could be used to predict pollutant emissions, to simulate autoignition with realistic fuels, to model the growth and oxidation of soot particles, and to model laboratory-scale turbulent combustion experiments in detail (3-d, sufficient chemistry). • Environmental Remediation and Processes: Petascale computing could be used to simulate the Hanford "leak event," probably used to simulate other regional ecological impacts requiring long-term, large-scale, 3-d, highresolution, 3-phase, multi-fluid flow and multi-component reactive transport, and to approach the elusive goal of real time multi-sensor data inversion. However, detailed scaling estimates are available only for the Hanford event. (One of the SCaLeS discipline coordinators notes a current lack of priority in accessing high-performance computer resources). 6/28/04 20 Applications Summary -Examples • Materials Science: Petascale computing could be used to advantage for problems such as high temperature superconductivity, magnetics, and toughening ceramics, but many applications (and PIs) in the field are focused more on capacity rather than capability. • Nanoscience: Petascale computing could be used to carry out molecular dynamics simulation of early key steps in the growth of colloidal quantum dots, the calculation of the electron transport properties of organic molecules, and the characterization of a 1000-atom FePt particle (perhaps applicable to future storage devices). Petascale Applications -Accelerators Impact of Petaflop-scale Computing: Application -Accelerators Accelerators Programmatic impact to be gained by access to capability Petaflop-scale computing Please indicate a few bullets which indicate the potential impact of Petaflop/s-scale computing as defined in the email cover letter. • Getting the most Science from the Nation's particle accelerators -using petascale modeling in concert with theory and experiments to optimize performance and expand operational envelopes • Improved designs for future accelerators -using petascale modeling to reduce cost & risk • Development of novel, groundbreaking methods for particle acceleration -using petascale modeling, in concert with theory and experiment, to explore, optimize, and implement laser-and plasma-based accelerators Major scientific challenges to be addressed Indicate the scientific challenges that are associated with the entries in the box above. • Optimizing the performance of an accelerator is an extremely challenging task: the beam behavior is governed by a combination of nonlinear effects and collective effects that can degrade beam quality and beam intensity and can lead to beam instabilities. Using petascale computing to improve accelerator performance will require a combination of petascale hardware and software resources (to perform and analyze the simulations), beam measurements, and mathematical methods for code validation, code calibration, uncertainty analysis, and prediction. An example is provided by the Large Hadron Collider, which is expected to come on line at the end of the decade. This is a multi-billion dollar facility, in which the US investment is approximately 1 billion dollars. When this machine comes on line, high-end computing will play an important role in commissioning, understanding beam behavior, and optimizing the accelerator performance. An important collective phenomenon known as the electron-cloud effect will be a key issue, and it is now being vigorously studied using terascale resources. A complete, high-fidelity simulation will require the use of near-petascale resources. • Accelerators are among the largest and most complex scientific instruments ever built, and future accelerators will "push the envelope" even further, particularly with regard to beam intensity. Because of their size, small changes in the design of large accelerator facilities can have huge financial consequences. "Over-designing" a machine (i.e. using an extremely conservative design) can cost hundreds of millions of dollars in capital costs; conversely, accelerator system optimization and better decision-making through high-fidelity, end-to-end petascale simulations can lead to designs that save hundreds of millions of dollars. • The successful development of ultra-high gradient accelerators through the laser or particle beam driven approach would have huge consequences for science, industry, and medicine. But, though experiments have already demonstrated gradients 100x to 1000x larger than conventional technology, it is extremely challenging to control and stage plasma sections into usable, production-capable particle accelerators. The systems themselves involve the simultaneous interaction of beams, plasmas, and radiation under extreme conditions, making diagnostics difficult. As a result, petascale simulations, used in concert with theory and experiment, provide one of the most powerful tools to understand these complex systems, and to ultimately design and implement plasmas-based accelerators. it is extremely challenging to control and stage plasma sections into usable, production-capable particle accelerators. The systems themselves involve the simultaneous interaction of beams, plasmas, and radiation under extreme conditions, making diagnostics difficult. As a result, petascale simulations, used in concert with theory and experiment, provide one of the most powerful tools to understand these complex systems, and to ultimately design and implement plasmas-based accelerators. What is the throughput (Tflops/s sustained) today on a single run of the longest calculations that are made? Please indicate the code efficiency and/or the computer peak performance. Impact of Petaflop-scale Computing: Application -Accelerators Accelerators Programmatic impact to be gained by access to capability Petaflop-scale computing Please indicate a few bullets which indicate the potential impact of Petaflop/s-scale computing as defined in the email cover letter. Projected increase in software efficiency? If you are counting on an increase from better algorithms (historically, algorithm improvements have approximately matched improvements in hardware), please indicate the factor you've used. • We are expecting a gain of at least ten fold in efficiency from our SciDAC efforts in computer science and applied mathematics to develop better algorithms More than 10 3 fold increase in problem size over a decade
doi:10.2172/919110 fatcat:j4vtreywhzazper5hte7bes5om