-
Notifications
You must be signed in to change notification settings - Fork 7
home
Numbox is the highly-parallel framework for engineering applications developed at IT4Innovations National Supercomputing Center (http://numbox.it4i.cz/). The main focus of the development team is to create highly efficient parallel tools suitable for parallel machines with tens or hundreds of thousands of cores. The added value of the framework is our
- mesh pre/post-processing tool mesio with the ability to load and store various sequential unstructured mesh formats in parallel, and
- massively parallel linear solver espreso based on FETI Domain Decomposition Methods, which is capable of solving problems with over a hundred billion unknowns using thousands of compute nodes.
Mesio is a highly-parallel loader and convertor of external unstructured mesh databases. It is composed of several commonly used algorithms that together provide a robust solution for the pre/post-processing phase of solving large-scale engineering problems (pdf). With mesio a user is able to use the same sequential database file with an arbitrary number of MPI processes. It allows engineers to use their favorite tools for the creation of numerical models without any penalty for a parallel run.
Mesio is composed of (i) the mesh builder that is able to reconstruct an unstructured mesh from randomly scattered data across parallel processes in a distributed memory of an HPC machine without gathering data into a single node and (ii) lightweight parallel parsers of given mesh databases. Hence, the usage is not limited to any particular format -- only a simple parallel parser has to be implemented for a new format. Currently, the following parsers are implemented:
- ANSYS CDB
- EnSight
- VTK Legacy
- XDMF
- Netgen
- OpenFOAM (partially)
- Abaqus (partially)
An output database stored by mesio is also in a sequential form for simple usage with a favorite visualization tool. The following formats are available:
- VTK Legacy
- XDMF
- EnSight
- STL surface
Espreso is a set of several highly-scalable solvers based on the methods of domain decomposition designed to take full advantage of today's most powerful petascale supercomputers (pdf). It contains in-house developed FETI based algorithms including the Hybrid Total FETI method suitable for parallel machines with tens or hundreds of thousands of cores.
The solver also contains a general API for usage in third party software. The API has been successfully used to connect espreso with the Elmer (pdf). Even though espreso is a C++ library, the API uses plain C only. Hence, it is easy to use it with various other languages such as Fortran.
Both mesio and espreso can be configured by ecf files. In addition, espreso can be also configured by a simple GUI.