GIVEN the sheer scale of modern supercomputers, it is tempting to assume that they are powerful enough to solve almost any problem imaginable. However, some of the computing challenges Laboratory researchers face are so complex that they compel us to persistently push computing’s boundaries. One of these challenges is maintaining a safe, secure, and effective nuclear stockpile in the absence of testing—an effort that relies in large part on supercomputers and the skilled computer scientists needed to perform sophisticated computer simulations.
As discussed in the article Reaching for New Computational Heights, Sequoia, the first IBM BlueGene/Q machine and one of the world’s fastest computers, was installed at Lawrence Livermore in 2012. With Sequoia, the Laboratory has demonstrated once again its leadership and expertise in designing, commissioning, and using high-performance computing (HPC) systems. BlueGene/Q was developed by IBM and Lawrence Livermore and Argonne national laboratories through an iterative and equitable design process. This process ensured that all three partners were involved in the decision making and that the resulting system could support both laboratories’ mission needs.
Since the start of the Accelerated Strategic Computing Initiative (ASCI)/Advanced Simulation and Computing (ASC) era, we have worked closely with IBM researchers for the delivery of five generations of Livermore supercomputers—ASCI Blue, ASCI White, ASC Purple, BlueGene/L, and now Sequoia. These projects have demonstrated that complex and ambitious computing technology development efforts are best accomplished through collaborations between government and industry. By sharing risk and expertise, we can achieve astonishing outcomes.
Now that Sequoia has been commissioned, tested, and tuned, we are eager to put it to work, executing sophisticated physics simulations and calculations for the National Nuclear Security Administration’s stockpile stewardship mission. With Sequoia, we will produce higher fidelity simulations with more accurate physics that will help us improve our existing models and codes. The machine will also allow us to routinely perform large sets of uncertainty quantification (UQ) calculations to better understand and reduce sources of uncertainty in our weapons codes. Working through the 100 to 200 potential failure modes in a nuclear weapon system, determining the margins of uncertainty for each, and then integrating and balancing these uncertainties across the system requires tens of thousands of simulations.
UQ can be challenging and computationally intensive, but in time, it could become an essential element of the design process for many large engineered objects, especially those that cannot easily be physically tested before use. For instance, UQ techniques were used during the development of BLU-129/B, a low-collateral-damage conventional munition codeveloped by Livermore for the U.S. Air Force. BLU-129/B also serves as a great example of how robust simulations can shorten design and development time. With supercomputing support, we moved from concept to field deployment in 18 months. Even accelerated Department of Defense design cycles typically take 4 to 6 years.
Supercomputing reduces the investment in time and testing needed to bring a product to market—not just for defense equipment but a full spectrum of high-quality goods. The industrial facet of computing matters because computational excellence is both a national and economic security issue, and countries such as China are offering intense competition on both fronts. Investing in computing and encouraging wider adoption of HPC simulation in product development is essential if our nation wishes to prevail in this competition.
For our part, we continue to support both national security and competitiveness through our computing. We routinely partner with companies and other laboratories to develop new software and systems that further our mission-centered science and engineering work and the field of computing itself. Vulcan, a smaller BlueGene/Q machine procured along with Sequoia, will be used to power industrial research through Livermore’s HPC Innovation Center. The Laboratory may hold an unmatched record of achievement in supercomputing, but we are not resting on our laurels.