Researchers from commercial property insurer FM Global use OLCF-developed middleware to drastically improve fire suppression simulations
Researchers from commercial and industrial property insurer FM Global want to improve fire protection standards across the country.
The company, which insures one in three Fortune 1000 companies with a unique focus on helping clients prevent losses, is dedicated to understanding how fires spread and, conversely, how they can be contained.
To that end FM Global researchers perform rigorous fire-spread and fire-suppression testing experiments at the company’s research campus in Rhode Island, which is home to one of the largest fire testing laboratories in the world—a football-field-sized fire testing facility with movable ceilings adjustable up to 60 feet in height. Experiments only go so far, though, and researchers have increasingly turned to computational modeling to understand the complex dynamics that contribute to fire spread.
FM Global’s interest in computationally modeling fires led it to a partnership with the US Department of Energy’s (DOE’s) Oak Ridge National Laboratory (ORNL) and its supercomputing facility, the Oak Ridge Leadership Computing Facility (OLCF). The OLCF, a DOE Office of Science User Facility, is home to the Cray XK7 Titan supercomputer.
While access to Titan, America’s fastest supercomputer, enabled FM Global researchers to increase their knowledge of fire spread dynamics and suppression methods, it was the opportunity to collaborate with computational experts that helped pave the way for future high-performance computing success.
“Having close working relationships with experts in the field is almost as important for success as use of a supercomputer as powerful as Titan,” said FM Global research scientist Yi Wang. Wang and fellow FM Global research scientist Karl Meredith worked with OLCF staff members to optimize their FireFOAM code. FireFOAM is a modified version of OpenFOAM, a widely used, open-source fluid dynamics code.
Researchers know that writing computer code to run simulations on the world’s biggest supercomputers is a challenge, but as computing resources have gotten larger and more powerful, it has become increasingly difficult to manage data and the flow of information in and out of a machine.
To improve input/output (I/O) for FireFOAM, the FM Global team partnered with OLCF staffer Norbert Podhorszki and Mark Olesen of ESI-OpenCFD, an organization that contributes to the development of OpenFOAM, to implement the OLCF’s Adaptable I/O System (ADIOS). ADIOS serves as middleware that researchers can use to quickly describe and organize datasets that need to be put on, or taken off, Titan. Implementing it successfully for FireFOAM also demonstrated that ADIOS could help improve I/O performance for OpenFOAM too, benefiting that code’s broad community of users.
Podhorszki pointed out that for codes such as FireFOAM and OpenFOAM, every object being calculated has its own I/O, meaning that for a simulation running on 1,000 Titan nodes, there are 1,000 I/O processes happening. “We turn off a code’s original I/O so that ADIOS can take those 1,000 processes and turn them into 1,” Podhorszki said.
As a result of this work, ESI-OpenCFD, FM Global, and ORNL now have an ongoing collaboration to add an ADIOS-based I/O interface into OpenFOAM+. “There is now a full-time developer at OpenCFD who is making a stable product from our early prototype, which was presented at the annual OpenFOAM workshop in 2015,” said Podhorszki. “And future development will focus on how to read data into ParaView, a widely used, open-source data analysis and visualization application. This is very important for all potential users.”
Meredith presented about this collaboration at the 11th annual OpenFOAM workshop, held June 26–30 in Guimarães, Portugal. “The major takeaways from my talk were that OpenFOAM I/O can be greatly improved, and ADIOS is an excellent roadmap to facilitate this improvement,” Meredith said. “Ultimately, ADIOS in OpenFOAM will allow OpenFOAM to be much more scalable on state-of-the-art high-performance computing machines.”
Oak Ridge National Laboratory is supported by the US Department of Energy’s Office of Science. The single largest supporter of basic research in the physical sciences in the United States, the Office of Science is working to address some of the most pressing challenges of our time. For more information, please visit science.energy.gov