Entries filed under “Applied HPC”

There are few issues today that capture as much attention as climate change: from policy makers and soccer moms to activists and mathematicians, a great deal of the world’s intellectual capability is focused on understanding, managing, and reversing the damaging climate effects of human activity on planet earth. The scope and complexity of the climate change problem pushes the boundaries of our scientific and engineering capabilities, even as the vast scale of the problem challenges our concepts of governance and the organization of civil society.

Climate scientists examine how elements such as atmosphere, land surface, ocean and sea ice systems interact to create global weather patterns whose statistics compiled over long periods of time define the climate state. They use sophisticated mathematical algorithms and complex calculations on supercomputers to gain a better understanding of climate system processes, and to use that understanding to predict changes in the Earth’s climate.

In fact, computer simulation is one of the few tools available to scientists and policy makers for understanding climate change. Numerical models allow us to validate our evolving understanding of the mechanisms of climate change, but the models are so complex that ordinary computers will not suffice. This is precisely the kind of challenge that supercomputers were designed to address, and supercomputing is a vital part of the climate science tool chest.

Climate Simulation at SC10

Climate science is a major focus area at this year’s Supercomputing Conference (SC10), a reflection of the focus that the HPC community is placing on this important issue. SC10, being held in New Orleans this year, will again attract scientists, researchers, students and businesses from around the world.

“The topic of climate has never been more relevant to a meeting like SC10 given the growing importance of simulation to addressing climate change questions from policy makers and stakeholders,” said James Hack, National Center for Computational Sciences director at Oak Ridge National Laboratory and the SC10 climate simulation thrust co-chair. “HPC will become an even more critical resource to help the broader research community to develop solutions to potential environmental impacts.”

During the conference, attendees will gather to learn from some of the world’s foremost experts in climate change and computation, and participate in exchanges that may shape future solutions. From masterworks sessions on the future of computing in climate science to papers on computational and data analytic issues, SC10 is a must-attend event for anyone working in the intersection of computing and the climate.

“This topic is particularly relevant to SC10, in view of recent extreme weather events, such as the very mild winter in Vancouver for the 2010 Olympics while Washington, D.C. experienced the heaviest snowfall in history. It is an open research question whether extreme weather events are related to climate change, ” said William Sawyer, computational scientist at the Swiss National Supercomputing Centre, and the SC10 climate simulation thrust area co-chair. “HPC is a critical resource to help the broader research community understand the connection between extreme weather events and climate, develop solutions to current climate issues, and help minimize the environmental impact of future events.”

Those who want to learn more will find a rich selection of papers, talks, and learning opportunities in New Orleans this year.

Learn from the masters

SC10 presents a rare opportunity to learn about the latest in climate science — and the latest thinking in the future of climate science — from the best in the world.

Terry Davies from the UK Meteorological Office will give a plenary talk on Thursday morning looking out over the next 20 years of climate prediction research, while the conference Masterworks program will provide even more opportunities to learn from the best in the world. Princeton’s V. Balaji, Richard Rood from the University of Michigan, James Kinter from the Institute of Global Environment and Society, and Mark Govett from NOAA will present during the week in separate sessions on topics ranging from computation and data management to new advances in our approach to science and the potential of GPUs in climate research applications.

Those looking for more interaction will enjoy Robert Jacob’s Birds of a Feather session on the analysis and visualization of large climate data sets, as well as the panel on “Pushing the Frontiers of Climate and Weather Models” with panelists Christiane Jablonowski, David Randall, Terry Davies, William Putman, Shian-Jiann Lin, and Peter Lauritzen.

And, of course, the week features many papers focused on computational technologies and techniques that those interested in climate science will want to keep in their toolbox.

Time to Dig In

The thrust areas at SC10 offer a unique opportunity to dig deep into critical issues driving the supercomputing community. Tune into events in the Climate Simulation thrust area throughout the week to be sure you are well-positioned to understand the issues of today, and to plan well for the science of tomorrow.

Linda Barney owns Barney and Associates, a technical, marketing writing and web firm in Beaverton, Oregon that provides writing and web content for the high tech, government, medical and scientific communities. Readers can reach her at linda@[email protected]

The staff and resources at TACC were recently part of an effort to add HPC-enabled calculations to the FAA’s operational safety workflow. Here is the outline of the full story which can be found on TACC’s website (complete with cool pics)

…if a crack is detected in an aircraft structure, does the problem affect only one plane, or is the failure systemic to the aircraft model or part? With lives and livelihoods on the line, officials must decide whether to institute a new inspection regime, keep the status quo, or ground the fleet.

Until recently, the FAA had little to go on when making this decision for small airplanes or the general aviation fleet. As a result, Harry Millwater, associate professor of mechanical engineering at The University of Texas at San Antonio (UTSA), working with post-doctoral researcher Gulshan Singh, and graduate student Juan Ocampo, developed a state-of-the-art structural integrity software, called SMART (SMall Aircraft Risk Technology), that in a matter of minutes can run thousands of simulations on a given part of a plane and provide a detailed report of the likelihood of a crack initiating in terms of both “hours and flights to failure”.

The invididual simulations aren’t long, but the FCC wanted to create an ensemble analysis that looks at a range of probable values for the critical variables in each situation. This can mean tens of thousands of runs that need to be completed rapidly in response to a detected failure in a flight system.

For that reason, Millwater approached TACC with the goal of “parallelizing” their code—making it capable of running simultaneously on multiple processing cores—and ultimately speeding up its performance so it could become a valuable, real-world tool.

…Working with Schulz, Millwater was able to make his code run 188 times faster by instituting a new MPI (Message Passing Interface) version that can efficiently distribute the calculations onto 256 processors (or the equivalent of more than 100 PCs).

“Something that took a couple of hours for analysis now took 42 seconds. You can’t beat that,” said Millwater.

The FAA has been using early versions of the code in exercises since last summer, and will take delivery of the final code at the end of this month when it will enter the FAA’s operational workflow.

The nice folks working on the Weather Research Forecast [WRF] code base announced a bug-fix release this afternoon.  Version 3.2.1 includes mostly bug fixes, but a few improvements.  They include:

• Fixes for producing wrffdda file
• Fixes for restart with auxiliary inputs (wrflowinput, wrffdda)
• Various fixes for physics, including URBPARM.TBL, MYJ PBL, YSU PBL, and RRTMG_LW
• tracer option without WRF-Chem
• curvature term correction for rotated lat-lon projection in ARW
• more namelist cross-checks

For more detailed updates for WRF-ARW, check out the UCAR website here.  For more info on the updates in WRD-NMM, check out the DTCenter website here.

…no word yet what its going to do with the furniture it wins.  According to an article in eWeek today, the IBM computer called ‘Watson’ [based on the BlueGene] will officially compete on the long-running television game show Jeopardy!.  The Blue-Gene-based machine will utilize technology called DeepQA developed by a research group at IBM Research.  Dr. David Ferrucci heads up the lab in charge of the project.

According to the source article, the official contest will occur sometime this Fall.  No additional details on the underlying hardware were provided in the article.

NCSA has posted another video briefing from its annual Private Sector Program Annual Meeting, this time about what Caterpillar does with HPC

At the recent NCSA Private Sector Program Annual Meeting, Keven Hofstetter described how Caterpillar uses high-performance computing to enhance product development and outlined a vision for extremely accurate, integrated simulation.

You can watch the movie here.

According to news out of TACC recently, Harvard Medical School researchers recently published results from computational studies they’ve been doing of a protein that is crucial to the hearing process

Cadherin-23 is also one of the proteins that malfunctions in individuals with hereditary deafness. Scientists believe one in 1,000 individuals in the U.S. are affected by this kind of disease, and 7.5 percent of hereditary deafness cases are caused by mutations in cadherin-23.

…The Harvard research team, including Sotomayor and Wilhelm Weihofen, and led by Rachelle Gaudet and David P. Corey, used the Ranger supercomputer at the Texas Advanced Computing Center (TACC) to simulate the 3D structure of cadherin-23 obtained from x-ray data. Then, they set the protein in motion, simulating the behavior of each atom as it reacted to forces that mimicked the effect of sound waves.

In simulations that involved up to 355,000 atoms, the team found that these particular proteins aren’t springy, but stiff. And they found a clue about what goes wrong in certain types of hearing loss

The simulations also showed that some deafness-related mutations to the protein do not alter the fold and strength of cadherin-23 directly. Rather, they change the way the protein binds calcium, which ends up weakening its structure.

Nifty application of HPC, and some very common codes (NAMD and VMD).

The gulf oil spill is (justifiably) receiving much of the world’s scientific and engineering attention, so it is no surprise that there are two HPC groups in the news right now working on the problem.

The University of Texas at Austin announced last week that researchers there are using Ranger — a Sun supercomputer that at one time was ranked 7th in the world and which today is still number 11 — to predict impact of oil spilling from the failed BP Deepwater Horizon oil well

With an emergency allocation of one million computing hours from the National Science Foundation TeraGrid project, the researchers are running high resolution models of the Louisiana coast to track the oil spill through the complex marshes, wetlands and channels in the area.

…”What our model can do that a lot of the other models can’t do is track the oil spill up into the marshes and wetlands, because we have fine-scale resolution in those areas,” he said.

This kind of detail will help the scientists determine how the oil may spread in environmentally sensitive areas. The team’s 2-D and 3-D coastal models also will take into account the Gulf of Mexico waves, which may bring the oil closer to the Texas coast.

The researchers are running 72-hour forecasts updated with satellite data about the current location of the spill at 50-meter resolution. UT claims that this is 10-20 times more refined than other simluations being conducted. Each run takes about 3 hours on 4,096 cores.

NCAR is also working together to produce simulations about a likely path for the spill

The computer simulations indicate that, once the oil in the uppermost ocean has become entrained in the Gulf of Mexico’s fast-moving Loop Current, it is likely to reach Florida’s Atlantic coast within weeks. It can then move north as far as about Cape Hatteras, North Carolina, with the Gulf Stream, before turning east. Whether the oil will be a thin film on the surface or mostly subsurface due to mixing in the uppermost region of the ocean is not known.

The code being used is POP, and simulations are being run at the New Mexico Computer Applications Center and at Oak Ridge National Laboratory.

Ohio State University systems biologist Dan Siegal-Gaskins is using HPC to help answer the question of how a cell knows what to turn into as it develops. To help him, he is studying the plant world’s answer to fruit flies, Arabidopsis thaliana (or mouse-ear cress)

At a specific phase of Arabidopsis leaf development, cells on the surface of the leaf receive genetic instructions to become either one of the majority ‘pavement’ cells or a large hair-like cell known as a trichome. The specific function of trichomes is unclear, although they may be involved in preventing infection, protecting delicate tissues on the underside of the leaf, or reducing the amount of water lost to evaporation.

…The mathematical model Siegal-Gaskins constructed consists of seven differential equations and twelve unknown factors. For his preliminary studies, he turned to OSC to choose random values for the unknowns and solve the equations for millions of different random value sets.

“Due to the large range of possible parameters and the complexity of the problem, we took advantage of OSC’s parallel processing capabilities and the MATLAB computing environment,” Siegal-Gaskins said. “This process was repeated for five million randomly-chosen parameter sets, and the set that gave us the closest agreement with experimentation was kept.”

It’s interesting that the team is using MATLAB for this problem — I’ve seen MATLAB function as a gateway into HPC for many teams of scientists that hadn’t previously used HPC before. I have a sense that its value in that role is seriously underestimated by mainstream HPC centers.

“Our bcMPI software, initially released in 2006, interfaces with HPC cluster technologies like PBS and Infiniband when executing MATLAB scripts on a cluster,” explained David Hudak, director of HPC engineering at OSC. “Over the last year, we have been working to improve the accessibility of parallel MATLAB. We designed Remote MATLAB Services (RMS) to enable our users to transition MATLAB scripts developed on their laptops to HPC resources. Dan was an early adopter of OSC RMS, and we learned a lot from his feedback. It was a very good fit for his needs.”

With the combination of computational modeling, literature-based analyses and laboratory experimentation, Siegal-Gaskins and Morohashi determined that the three cell fate proteins seem to constitute an “incoherent feed forward loop,” a relationship in which a master regulator (MR) triggers expression of two genes involved in the initiation of trichome cell development, one of which (G1) later suppresses expression of the other (G2).

More in the full story from OSC.