ASCR Monthly Computing News Report - January 2008
The monthly survey of computing news of interest to ASCR is compiled by Jon Bashor (JBashor@lbl.gov) with news provided by ASCR Program Managers and Argonne, Fermi, Lawrence Berkeley, Lawrence Livermore, Los Alamos, Oak Ridge, Pacific Northwest and Sandia National labs. Contact information and links to additional information, where available, are included with each article.
In this issue...
DOE Awards 265 Million Hours of Supercomputing Time to INCITE Projects
DOE’s Office of Science announced January 17 that 265 million processor-hours were awarded to 55 scientific projects, the largest supercomputing resource awards in the Department’s history and three times that of last year’s award. The allocations of supercomputing and data storage resources will be made under DOE’s Innovative and Novel Computational Impact on Theory and Experiment (INCITE) program, which supports computationally intensive, large-scale research projects.
The projects—with applications from aeronautics to astrophysics, and from climate change to combustion research—were chosen based on their potential breakthroughs in the science and engineering research and their suitability for using supercomputers. These awards will allow cutting-edge research to be carried out in weeks or months, rather than years or decades, giving scientists access to some of the world’s most powerful supercomputers at DOE national laboratories.
The 2008 projects were awarded time at DOE’s Leadership Computing Facilities at Oak Ridge National Laboratory in Tennessee and Argonne National Laboratory in Illinois, as well as at the National Energy Research Scientific Computing Center (NERSC) at Lawrence Berkeley National Laboratory in California, and the Molecular Science Computing Facility at Pacific Northwest National Laboratory in Washington. Of the 31 new projects and 24 renewal projects selected, eight are from industry, 17 from universities and 20 from DOE labs as well as other public, private and international researchers.
Zoltan and PETSc Go with the Flow with Help from SciDAC
The PETSc toolkit now supports the Zoltan parallel partitioning and load balancing toolkit as an external library. Zoltan’s hypergraph partitioner can be used to partition meshes, thereby improving load balance and reducing communication in large-scale, parallel applications using unstructured meshes in PETSc. This work, led by Matthew Knepley (ANL), was a SciDAC collaboration between the TOPS Center and the CSCAPES Institute. The PETSc-Zoltan link will soon be used by the PFLOTRAN reservoir simulation code in the “Going with the Flow” groundwater SciDAC project.
“Berkeley View” of Multicore Challenges Lauded by HPCwire
A report co-authored by Kathy Yelick and John Shalf at NERSC was among the top issues and trends singled out by the editor of HPCwire for its last issue in 2007. Editor Michael Feldman wrote: “The Landscape of Parallel Computing Research: The View from Berkeley became a wake-up call to the computing community about the perils and pitfalls of our manycore destiny. Was anyone listening? Maybe. In the past year, both Intel and Microsoft spent a gazillion dollars for parallel computing R&D and education. Universities like Purdue, LSU, the University of Manchester, MIT and many others are expanding their HPC curriculums for the next crop of students. By the time these kids start to graduate in 2010, the manycore chips will be spilling out of the fabs.” Read the article at this link
Sandia Researchers Rehabilitate Quadrilateral Finite Elements
An important class of finite element methods often used to model porous-media flows and semiconductor devices suffers from reduced performance on unstructured quadrilateral grids that can, in severe cases, result in complete loss of convergence. Sandia researchers have developed a reformulation that restores convergence rates of these methods where the key idea is to replace the analytic divergence operator by its natural mimetic approximation. Theoretical analysis of reformulated methods shows that they are optimally convergent, and conversion of existing codes to reformulated methods is trivial, making the approach much more attractive than retrofitting codes with new element definitions.
Argonne SiCortex Helps Develop Library for Computing Properties of Carbon-12
Computer scientists Ewing Lusk (Argonne National Laboratory) and Ralph Butler (Middle Tennessee State University) and physicist Steven Pieper (Argonne) have been using the 5,832-processor SiCortex system at Argonne to develop an Automatic Dynamic Load Balancing (ADLB) library for use in SciDAC’s Universal Nuclear Energy Density Function (UNEDF) project. One of the aspects of the UNEDF work will be to use the Argonne Green’s Function Monte Carlo (GFMC) program to compute properties of carbon-12 using Argonne’s Blue Gene/P.
To date, the GFMC program has been used on 2,000 processors. To be able to use the 130,000 processors of the full BlueGene/P — an order of magnitude more — a new parallelization is needed. Initial work on this new scheme is being done on the SiCortex. In the first attempts using 1,600 processors, the job ran out of memory. Thanks to a close collaboration between Argonne and SiCortex, however, Butler was able to learn about unadvertised debugging hooks in SiCortex’s implementation of MPI. These clearly showed the cause of the excessive memory use, and changes were made to ADLB. The results have been impressive: the code has been used on up to 5,400 processors, with essentially perfect scaling up to 4,400 processors. The next step is to achieve perfect scaling up to the 5,600 processors that can be used for a job on the SiCortex. At that point, development work will continue on the BG/P.
New State of the Art Established in Solving Production Scheduling Problems
Under ASCR-AMR funding, Jean-Paul Watson (SNL) and J. Christopher Beck (University of Toronto) recently developed an optimization algorithm that establishes a new state-of-the-art performance benchmark for the classical Job Shop Scheduling Problem or JSP. The JSP is the dominant scheduling model analyzed by the operations research community and is among the most difficult NP-hard problems to solve in practice. The new algorithm represents a novel hybridization of tabu search and constraint programming that identifies twelve new, best-known solutions to a broadly studied set of forty JSP benchmark problems and is able to consistently generate high-quality solutions - a major issue with many competing algorithms.
NOAA Team Uses NERSC to Produce Finely Detailed Climate Model Simulations
A team of climate researchers who obtained early access to NERSC's new Cray XT4 said the powerful system produced simulations that offered details of oceanic and atmospheric phenomena, results that were difficult to obtain from other supercomputers before. The climate modeling project, called the Coupled High-Resolution Modeling of the Earth System (CHiMES), began as a collaboration between the National Oceanic and Atmospheric Administration's Geophysical Fluid Dynamics Laboratory (NOAA/GFDL) and DOE. At DOE's behest, scientists from GFDL proposed a set of experiments using climate models with resolutions many times higher than those in the standard models. GFDL scientists, located in Princeton, N.J., had developed models capable of modeling the global atmosphere at resolutions down to 5 km, and the ocean at resolutions between 10 km and 20 km (compared to 100 km for many models). They also have designed experiments which generated one to four terabytes of data for every year of simulation. NERSC provided GFDL with the computation resources for this challenge by setting aside over 800,000 CPU hours on the new Cray XT4 named Franklin.
Simulation Aids Development of First Coal Plants with Near-Zero Emissions
DOE researchers at the National Energy Technology Laboratory (NETL) and Oak Ridge National Laboratory (ORNL) are using supercomputers to improve the industrial design of facilities that tap coal's oxidative potential while trapping its pollutants. The highest-resolution coal-gasification simulation to date should enable faster development of clean coal technology that emits nearly no nitrogen oxide, mercury or sulfur and that sequesters most carbon dioxide.
NETL researchers will run simulations on Jaguar, ORNL's Cray XT4, for two weeks in January 2008 that will provide 1 millimeter resolution. Jaguar's high resolution means combustion can be modeled in great detail, enabling simulations that represent reality with unprecedented fidelity. Jaguar's enormous memory and speed produce quicker answers to scientific and engineering questions and accelerate development of the technology.
James J. Hack Named Director of NCCS at Oak Ridge
James J. (“Jim”) Hack has been selected to direct the National Center for Computational Sciences (NCCS), a leadership computing facility at Oak Ridge National Laboratory (ORNL), and to head ORNL’s Climate Change Initiative. As director of the NCCS, in partnership with the project director of the Leadership Computing Facility at ORNL, Hack will lead the high performance computing. As leader of the Climate Change Initiative, he will develop laboratory-wide programs in climate change and lead a team of scientists and engineers across ORNL in advancing the state of the art in Earth system discovery and policy through enhanced scientific understanding, Earth system modeling, and advances in computational and observational programs.
Hack joins ORNL from the National Center for Atmospheric Research (NCAR) with a distinguished career in climate science research. He has long collaborated with the international climate community and serves on the Department of Energy’s Advanced Scientific Computing Advisory Committee and the Working Group on Numerical Experimentation, which is supported by the Joint Scientific Committee for the World Climate Research Program and the World Meteorological Organization Committee for Atmospheric Sciences.
LBNL's Arie Shoshani Invited to Contribute to Database Encyclopedia
Arie Shoshani, head of SciDAC's Scientific Data Management Center and leader of LBNL's Scientific Data Management Group, has written two essay-type entries for the Encyclopedia of Database Systems, to be published later this year by Springer Verlag. According to the publisher, "The Encyclopedia of Database Systems is designed to meet the needs of research scientists, professors and graduate-level students in computer science and engineering.. Topics for the encyclopedia were selected by a distinguished international advisory board, and written by world class experts in the field." Based on his work in the field, Shoshani was invited to write entries on "Logical Models of Temporal Data" and "Summarizability Properties of Statistical Databases."
Ali Pinar of LBNL Elected Officer of SIAM Activity Group on Supercomputing
Ali Pinar, a researcher in the Scientific Computing Group, has been elected to serve as secretary for the SIAM Activity Group on Supercomputing (SIAG/SC). His two-year term began on January 1. Pinar's research focuses on combinatorial scientific computing, and he is particularly interested in tackling combinatorial problems that are directly associated with scientific and engineering goals. He has worked on vulnerability analysis of the electric power grid, interconnection networks for ultra-scale systems, energy-efficient disk systems, and supernova spectra analyses. SIAM is the Society for Industrial and Applied Mathematics
. The SIAM Activity Group on Supercomputing provides a forum for computational mathematicians, computer scientists, computer architects and computational scientists to exchange ideas on mathematical algorithms and computer architecture needed for high performance computer systems. The activity group promotes the exchange of ideas by focusing on the interplay of analytical methods, numerical analysis and efficient computation.
Argonne Leadership Computing Facility Accepts 111-TF Blue Gene/P System
On December 27, 2007, the Argonne Leadership Computing Facility (ALCF) officially accepted the 111-teraflops IBM Blue Gene/P, named Endeavour, as well as the test and development system Surveyor, and associated file systems. With the official acceptance, the center will begin transitioning the systems into production. During this time, a set of seven Early Science projects will be given access the supercomputer to help fully test the Endeavour system in a production environment while completing key science objectives.
Grassroots Cyber Security Town Hall Meeting Set for February
A "town hall" meeting to develop an R&D roadmap for cyber security to address long-term needs of the Department of Energy will be held February 11-13 at Argonne National Laboratory. Registration is open to anyone interested in the topic. A draft white paper will be developed into a report as a product of the workshop. For more information and to register, go to this link: https://wiki.anl.gov/cybercommunity
Advanced Scientific Computing Advisory Committee to Meet Feb. 26-27
The next Advanced Scientific Computing Advisory Committee (ASCAC) meeting will be held at the American Geophysical Union (1000 Florida Ave N.W., Washington, D.C.) on February 26-27. The meeting will be open to the public and include talks by Under Secretary for Science, Raymond L. Orbach; Associate Director of the Office of Science for ASCR Michael Strayer; and a dinner talk by Tony Hey of Microsoft Research. There will also be technical talks from Kathy Yelick (NERSC/LBNL on programming models), Bart Miller (University of Wisconsin on infrastructure scaling for performance), Mike Heroux (Sandia National Laboratories on parallel math libraries and algorithms), Bob Lucas (University of Southern California/Information Sciences Institute on the DARPA Exascale Study), and Deborah Frincke (Pacific Northwest National Laboratory on community planning for cyber security).
ASCAC was established in August 1999 to provide valuable, independent advice to the Department of Energy on a variety of complex scientific and technical issues related to its Advanced Scientific Computing Research program.
ALCF Offers INCITE and BG/P Workshop Series
The Argonne Leadership Computing Facility (ALCF) is organizing a series of INCITE and BG/P workshops in the coming months, starting with an "Introduction to BG/P" Workshop on February 20. This will be followed by an INCITE "Getting Started Workshop" on March 4-5 and a joint ALCF and Blue Gene Consortium Workshop on March 6 focusing on "BG/P Porting and Tuning." For further details, watch the ALCF website (www.alcf.anl.gov
Upcoming SciDAC Workshop on Combinatorial Scientific Computing
The CSCAPES SciDAC institute will host a workshop on combinatorial scientific computing and applications in Santa Fe, New Mexico, on June 10-13, 2008. The objective of this workshop is to bring SciDAC applications, centers, and institutes together with CSCAPES researchers to address combinatorial issues encountered in tera- and petascale computing. The workshop is organized around theme areas that include partitioning (load balancing), automatic differentiation, and parallel graph and sparse matrix algorithms.
LBNL Helps Organize Microbial Genomics and Metagenomics Workshops
The Biological Data Management and Technology Center (BDMTC) in Berkeley Lab's Computational Research Division will take part in three Microbial Genomics and Metagenomics Workshops this year at the DOE Joint Genome Institute (JGI). The five-day workshops will provide training in microbial genome and metagenome analysis. The workshops will offer two days of intensive seminars and three days of hands-on tutorials on a variety of data analysis tools, in particular the tools provided by the Integrated Microbial Genomes (IMG) family of systems developed by BDMTC in collaboration with the Genome Biology Program (GBP) at JGI. Nikos Kyrpides, head of GBP, David Gilbert, head of public relations for JGI, and Victor Markowitz, head of the BDMTC, are the workshop organizers. The first workshop was held January 7-11, and subsequent workshops will be held May 19-23 and September 15-19, 2008. Learn more about the workshops at http://www.jgi.doe.gov/meetings/mgm
PNNL to Host Summer School in Multiscale Mathematics and HPC
The third annual workshop in Multiscale Mathematics and High Performance Computing will be held July 14-17, 2008, at Pacific Northwest National Laboratory. The summer school, supported by DOE's Office of Advanced Scientific Computing Research, is a week-long workshop that equips researchers to advance understanding of complex physical processes occurring on a wide range of interacting length and time scales. Aimed at graduate and post-doctoral students in mathematics, scientific computing, materials sciences, geophysics, computational physics, and mechanical engineering, the workshop provides introductions to the mathematical and computational methods commonly used to model physical systems at various scales: continuum methods, discrete methods, statistical methods, network methods. Paul Whitney, Associate Division Director for Computational Mathematics at PNNL, is organizing the program.
NERSC to Host Cray Technical Workshop Feb. 26-27 in San Francisco
Registration is now open for the Cray Technical Workshop North America to be held Feb. 26-27, 2008 at the Hotel Nikko in San Francisco. Hosted by NERSC, the workshop will feature presentations from users and Cray experts on the latest scientific advancements using Cray supercomputers, as well as advice on how to achieve the optimum performance on Cray systems. Featured speakers include NERSC's Bill Kramer, Zhengji Zhao, and Katie Anytpas, Berkeley Lab's Terry Ligocki and Julian Borrill, as well as NERSC users Zhihong Lin (UC Irvine) and Nick Wright (SDSC). There will be additional user talks on performance and various vendor presentations on state-of-the-art subjects, especially related to quad-core usage and performance. For more information, see http://www.cray.com/workshop/northamerica