Computational Science Research Software for Manufacturing Simulations: Software Infrastructure Challenges
David Hudak, Ohio Supercomputer Center
Manufacturing competitiveness is a critical U.S. national interest that can benefit greatly from advanced virtual prototyping supported by high-performance computing (HPC) and related cyberinfrastructure (CI). Modeling and simulation provides a competitive edge to businesses by allowing creation of virtual prototypes at much lower expense, examination of more design alternatives and reduced time to market for new or revised products. We seek to spur computational science and CI advancements through the investigation of research areas (such as parallel solvers and CI technologies) whose solutions directly support manufacturing applications.
We examine issues that arise in the creation of a modeling application for metal-ceramic composite materials. This application models the displacement chemical reactions at the solid/fluid interface, in which sacrificial oxides are reduced by a molten metal and subsequently form a ceramic/metal composite. Supporting this simulation led to the consideration of a related open-source solver for Dissipative Particle Dynamics (DPD). The sequential code is approximately 10K lines of code in C developed under Visual Studio. While extending the DPD solver to HPC scale, we identified three major software challenges: (1) scaling open-source scientific codes, (2) leveraging new hardware technology and (3) improving developer productivity. Scaling code was achieved via a combination of the NSF-funded MVAPICH2 software and the Global Arrays Toolkit. The Intel MIC architecture was targeted as a representative new hardware technology and evaluated using proprietary tools. The NSF-funded Eclipse Parallel Tools Platform (PTP) and related HPC-center support was developed to improve productivity.
These activities demonstrate that collaboration between industry and computational science researchers can foster both competitive companies and compelling research. We believe that industrial adoption of advanced modeling and simulation is inhibited largely due to a lack of reliable, scalable and extensible software. Continued development of NSF software infrastructure is critical to address the priority to “Further U.S. economic competitiveness” described in NSF’s “Investing in America’s Future Strategic Plan 2006-2011”.