The CEA/NNSA collaboration currently focusses on four technical topics. In the sections below each of these areas is briefly described along with some technical highlights and pointers to additional information (e.g. open source code, documents, etc.). Technical contacts (CEA & NNSA) are also provided for each area.
This collaboration area focuses on visualization and data analysis and centers on the open source development of the Visualization Toolkit (VTK), which forms the basis of our customized visualization and analysis tools (ParaView, VisIt and LOVE). We are partnering on common VTK needs and working on methods for visualizing and analyzing data on advanced architectures that are memory and IO constrained.
1) Portions of the CEA visualization tool LOVE were contributed as Open Source Plugins to ParaView.
2) Multi-resolution display capabilities were added to multiple of our customized visualized tools allowing users to visualize large data sets on small computational resources.
3) In situ visualization and analysis capabilities were delivered to users throughout NNSA/ASC and CEA/DAM.
4) Several core algorithms were implemented for GPGPU architectures. A number of papers were written describing the results.
Links and References
More information can be found about VTK at http://www.vtk.org
More information can be found about ParaView at http://www.paraview.org/
More information can be found about VisIt at http://visit.llnl.gov/
Acceleration Hardware and APIs
This work explores areas of common interest in programming models for the coming exascale era. Initially, the intent was to aggregate NNSA and CEA lab's work in this area into a common repository.
Early work will focus on low-level APIs, micro-benchmarks, and mini-applications. Work in the out-years will include APIs and languages that provide higher levels of abstraction.
1) Los Alamos rewrote, and posted as open source, a CUDA version of the CEA mini-app: codehydro.
2) Sandia published technical report analyzing mini-app and their representation of characteristics of actual production codes.
3) CEA developed and published as open source the Hydro miniapp (in various forms such as Cuda, OpenCL, OpenMP…).
Links and References
The Los Alamos GitHub site is at http://github.com/losalamos
The Sandia Mantevo mini-apps are at http://mantevo.org
CEA’s open source repository is at http://github.com/HydroBench/Hydro.git
This collaboration focuses on research and development of parallel file system interfaces and tools, system resource management capabilities, operating system evaluation and tools, and software for high-performance interconnects. Current activities include the test and evaluation of technologies and tools associated with the Lustre parallel file system, the development and analysis of middleware to encapsulate application I/O requirements and abstract the capabilities of the underlying parallel file system, enhancements to the SLURM resource manager to support evolving extreme-scale application workflow requirements, evaluation of the Portals low-level interconnect programming interface, and evaluation of the Kitten lightweight kernel operating system.
- Lustre : ZFS backend development by LLNL tested by CEA, discussions and shared experiments about failover in Lustre for example can be considered as a success in that domain. Moreover presentations or focus on some tools like Robinhood or shine for example are also very interesting to highlight the filesystem administration.
- I/O layers : discussions, strategies, point of vue exchanges (diod versus ganesha and 9P for example), but also several group meetings about application IO layers like Hercules, PLFS, SCR have been done. This is a real success for this new collaboration subject.
- Slurm : SLURM workshop organization, new features like topology aware, kerberos support, high availability was some examples of the success of the collaboration (all these new features were integrated in the main branch of the SLURM opensource project). Moreover the link created between teams (exchanges through a mailling lists), allow the group to share experiences about more general things like batch or scheduling strategies.
- Operating system : the last meeting showed us that it would be very useful to collaborate on this subject especially for testing and exchanging about new OS (like kitten for example) and administration tools evaluations and enhancements (like the opensource nodediag).
Links and References
More information can be found about Lustre at http://www.lustre.org
More information can be found about SLURM at https://computing.llnl.gov/linux/slurm/
More information can be found about Portals at http://www.cs.sandia.gov/Portals
More information can be found about Kitten at http://software.sandia.gov/trac/kitten
More information can be found about Ganesha at http://sourceforge.net/apps/trac/nfs-ganesha/
More information can be found about Robinhood at http://sourceforge.net/projects/robinhood/
This work focuses on automatic algorithms for 3-D mesh generation and their extension to scalable parallel environments. The objective is to explore new theoretical advancements and develop new algorithms in mesh generation that will lead to practical advances in automatic meshing with an emphasis on all-hexahedral mesh generation technology.
The development of technologies to facilitate model preparation for analysis such as CAD support for meshing and geometry simplification are also within the scope of this joint work.
We have established a working technical collaboration for the exchange of ideas in the area of mesh generation research. This has included several short stays of 1 to 2 weeks of visiting technical staff to the US and France where new ideas have been explored and advanced. In addition, several peer reviewed technical papers co-authored by CEA and NNSA technical staff have been published. Co-Sponsorship of international conferences, such as the October 2011 Paris, International Meshing Roundtable, involved both CEA and NNSA and helped foster exchange of ideas to a broad international audience.
Links and References
Programs at NNSA and CEA laboratories continue to sponsor and advance technical work that leverages ideas developed through this collaboration.
The CUBIT Geometry and Meshing Toolkit (http://cubit.sandia.gov), a product of Sandia National Laboratories, is one example.
[Kow12] N. Kowalski, F. Ledoux, M. L. Staten and S. J. Owen, Fun Sheet Matching: Towards Automatic Generation of Hexahedral Block-Structured Meshes, Engineering with Computers, volume 28, issue 3, pages 241-253, July 2012.
[Sta10] M. Staten, J. Shepherd, F. Ledoux and K. Shimada, Hexahedral Mesh Matching: Converting Non-Conforming Interfaces into Conforming Interfaces, International Journal for Numerical Methods in Engineering, volume 82(12), pages 1475-1509, 2010.