Pentagon Supercomputers Puzzle Out How to Safely Airlift Coronavirus Patients
And that's just one of the various pandemic-related problems that Defense Department supercomputers are chewing on.
The Defense Department’s High Performance Computing Modernization Program, or HPCMP, has been in operation for nearly two decades, delivering supercomputing capabilities and computational science expertise to advance the agency’s mission and help solve some of its most crucial challenges.
Defense officials are now turning HPCMP’s supercomputing resources to overcome one of the most daunting challenges of the century: defeating the novel coronavirus pandemic.
“COVID-19 response is a whole-of-government effort,” HPCMP’s Acting Deputy Director Kevin Newmeyer told Nextgov Thursday. “The White House, [Defense] and our leadership in the Corps of Engineers view our role as one of supporting the efforts of scientists and researchers to meet the challenges facing the nation—be that airlift of COVID-19 patients, vaccine development, emergency hospital treatment facilities, or other challenges.”
The program usually provides more than 7 billion computer core hours annually to the Pentagon to support research, development and testing of military systems and platforms. Newmeyer said its computers are built to meet a range of defense-related workloads quickly and efficiently, and the present goal is to harness those and other HPCMP assets to accelerate and strengthen America’s response to the global health emergency.
“The value of supercomputers is that we can dedicate the equivalent of tens of thousands of individual computers to work on a single problem,” he explained. “By dividing the problem into smaller chunks or running the identical task across thousands of possible input choices, we can deliver results much faster with great accuracy.”
Managed by the U.S. Army Engineer Research and Development Center, or ERDC, HPCMP’s pivot to coronavirus-fighting efforts was partially sparked when the White House Office of Science and Technology Policy, Energy Department and tech giant IBM launched the COVID-19 High Performance Computing Consortium in late March. At that point, the HPCMP offered the consortium its open research system—Copper—a 15000 core Cray supercomputer, as well as access to its cadre of computational scientists.
“Simultaneously, we reached out to our service [high performance computing] leads offering to support [Defense] researchers working on COVID tasks priority access to our supercomputers,” Newmeyer said. “Traditionally, our computers run 24/7/365. We have taken steps in the workflow process to ensure that COVID-19 jobs run as soon as possible to get the results quickly.”
Newmeyer offered a peek into how the program’s assets and resources are supporting the fight against COVID-19.
Assessing the Risks of Transferring Patients
Prompted by an urgent operational request from the Commander of U.S. Transportation Command, Defense researchers are running supercomputers at the Air Force Research Laboratory’s Defense Department Shared Resource Center in Ohio—and leveraging efforts on software developed and acquired by HPCMP—to assess methods for reducing risk to aircrews and medical attendants in the event of airlifting COVID-19 passengers using the Air Force. Through one recent effort, the program’s experts launched computational fluid dynamics studies of airflow and droplets through an aircraft's interior.
Newmeyer explained that the commander “is looking at the potential for having to move COVID-19 patients to treatment facilities around the country,” and once the Air Force received a task to evaluate the dangers that would pose to aircrews and medical personnel, officials subsequently requested priority computer support from HPCMP.
“While the Air Force has long had the mission to move wounded troops to high level medical facilities, COVID-19 patients represent a different challenge due to the potential for airborne transition of the disease,” Newmeyer explained. “The airflow study enables mission planners to visualize how the air and any entrained droplets would move inside the plane’s flight deck and cargo bay so they can implement measures to control the movement of droplets and best mitigate the risks to the airmen.”
Observing airflows is a classic supercomputer problem in the field of computational fluid dynamics, Newmeyer noted, and officials used the program’s resources to conduct a droplets analysis that simulates movement from the inside of a C-17 Globemaster III aircraft. Developed over the course of 10 years, HPCMP’s use of specialized software on modern supercomputers is part of the program’s Computational Research and Engineering Acquisition Tools and Environments, or CREATE, software development effort. Newmeyer said the code they tapped into applies to a wide range of problems in aerodynamics because it’s physics-based.
“It breaks the computational space—in this case the interior of a C-17 transport plane—into discrete cells and steps very slowly thru time to provide a history of how the droplet disperses inside the aircraft—cubic inch by cubic inch over several minutes.,” Newmeyer said. “So you need a large computer because of the size of the problem, but also one that is fast so you can get results quickly.”
In this case, breaking the problem into pieces and sending it to a large number of processors that all communicate together to solve it sparks the gains in speed. As he explained it, “if you have a problem that takes 10 hours to run on 10 processors,” and then opt to “use 10,000, you can get the problem done in [one one thousandth] of the time.” That’s essentially a reduction of 10 hours to 36 seconds, which means officials saw a relatively rapid outcome.
“We received the request for support on March 27 and by last Friday the results were available,” Newmeyer said.
Research Supporting Vaccine Candidates
Other work underway encompasses a task from the Army Medical Command and the Walter Reed Army Research Institute, which is sponsoring an expert at the Southwest Research Institute, or SWRI, to conduct virtual drug screening of COVID19 vaccine candidates that can possibly be used against the virus. Though the resources available at SWRI only enabled about two million candidates to be investigated over a three-week period, HPCMP is offering up high-performance computing assets at the Army Research Lab at the Aberdeen Proving Ground in Maryland and the ERDC’s Defense Supercomputing Resource Centers to accelerate the computational studies.
The program is providing supercomputer expertise and 1 million core hours of computer time.
“The Southwest Research Institute problem being worked for the Army’s Walter Reed Institute is slightly different in that a large number of simulations need to be run quickly,” Newmeyer explained. “By spreading the individual simulations across thousands of computer cores, thousands of experiments can be simulated at the same time, significantly reducing the total time to investigate candidate compounds.”
He said that the SWRI researcher uses a computer program to virtually assess potential drug combinations and vaccines candidates against COVID-19. Computer scientists from the HPCMP worked directly with that researcher to “transport his code from a workstation application to one optimized for the supercomputer environment and architecture,” which could significantly speed up potential outcomes—and allow researchers to allot more concentration on the candidates that hold the most potential.
“Using the HPCMP supercomputers at ERDC and Aberdeen, it is possible to assess tens of millions of compounds in a week,” Newmeyer said. “We expect it will evaluate 40 million candidates in a week,” which is much higher than the 2 million candidates that SWRI could originally investigate over a three-week period
COVID-19 Modeling to Help FEMA
Through its parent company ERDC, the program has also unleashed efforts to support the Federal Emergency Management Agency with COVID-19 modeling.
While a range of modeling projects are ongoing, Newmeyer highlighted HPCMP computer experts work with researchers from ERDC’s Environmental Laboratory to adapt and improve the Susceptible, Exposed, Infectious, Recovered or ‘SEIR’ model that projects how COVID-19 infections would spread in a given community.
Much like the previously mentioned SWRI example, while this code was originally designed to run on a desktop computer, ERDC scientists and computational experts adjusted it to run on the Onyx supercomputer.
“This enables it to run much faster and to examine a range of possible assumptions on different variables such as population density and mobility to better inform FEMA and our partners,” Newmeyer said.
The acting deputy director also made a point to note that all of the researchers can access the HPCMP supercomputers remotely.
“Our usage has remained very high throughout this period of social distancing and telework,” Newmeyer said. He added that HPCMP’s “team remains shoulder-to-shoulder with our government, military, industry and academic partners to address current and future support requirements to defeat the virus.”