Interoperable Technologies for Advanced Petascale Simulations (ITAPS)

Abstract

Efforts during the past year have contributed to the continued development of the ITAPS interfaces and services as well as specific efforts to support ITAPS applications. The ITAPS interface efforts have two components. The first is working with the ITAPS team on improving the ITAPS software infrastructure and level of compliance of our implementations of ITAPS interfaces (iMesh, iMeshP, iRel and iGeom). The second is being involved with the discussions on the design of the iField fields interface. Efforts to move the ITAPS technologies to petascale computers has identified a number of key technical developments that are required to effectively execute the ITAPS interfaces and services. Research to address these parallel method developments has been a major emphasis of the RPI’s team efforts over the past year. Efforts to move the ITAPS technologies to petascale computers has identified a number of key technical developments that are required to effectively execute the ITAPS interfaces and services. Research to address these parallel method developments has been a major emphasis of the RPI’s team efforts over the past year. The development of parallel unstructured mesh methods has considered the need to scale unstructured mesh solves to massively parallel computers. These efforts, summarized in section 2.1 show that with the addition of the ITAPS procedures described in sections 2.2 and 2.3 we are able to obtain excellent strong scaling with our unstructured mesh CFD code on up to 294,912 cores of IBM Blue Gene/P which is the highest core count machine available. The ITAPS developments that have contributed to the scaling and performance of PHASTA include an iterative migration algorithm to improve the combined region and vertex balance of the mesh partition, which increases scalability, and mesh data reordering, which improves computational performance. The other developments are associated with the further development of the ITAPS parallel unstructured mesh adaptation procedures. Specific developments include: • Parallel boundary layer mesh adaptation integrated with parallel anisotropic mesh adaptation (section 2.4.1). • A new more scalable message packing library (section 2.4.2). • Support of periodic boundary conditions (section 2.4.3). We have continued to work closely with both the accelerator applications for COMPASS and fusion application for CEMM. For COMPASS, efforts have focused on providing specific unstructured mesh adaptation tools to deal with curved elements and mesh adaptation. For CEEM, we are working to provide the structures and methods needed for the M3D-C1 to go to full three dimensional configurations.