Incremental Generation of Polycubes Driven by Geometric Quality
The proposed work consists of designing and developing a new algorithm for generating block-structured hexahedral meshes using a “Polycube”-typed strategy. Usually, these methods deform a geometric domain G to be discretized into a polycube P, i.e. a polyhedron whose all the faces are orthogonal to one of the X, Y, or Z axes. This polyhedron can be easily discretized using a hexahedral mesh, which is then transformed via the inverse deformation to pave G.
Unlike traditional approaches, which are based on constructing the deformation function, we focus on the inverse function by treating the pair (geometry G, polycube P) as input. Our goal is to compute the inverse function F that transforms P into G, and, depending on some local properties of F, modify P to provide a “better” mesh of G from a geometric point of view.
Generation of Curved Hexahedral-Dominant Block Structures Using Median Objects
The automatic generation of block-structured hexahedral meshes is a challenging problem, solved in practice through the use of dedicated interactive software, such as Magix3D, which is developed and maintained at CEA DAM. Using such software, a specialist engineer may spend several days creating the expected mesh.
The objective of this thesis is to help such an engineer to quickly sketch initial block structures with a hexahedral-dominant structure by relying on the medial object of the 3D geometric domain to be discretized. When considering CAD-type geometric domains, represented by their boundary, the medial object provides volumetric information intrinsic to the domains, which will guide the engineer in creating individual hexahedral blocks.
The objective of this thesis is twofold: first, to propose an efficient method for generating 3D medial objects; and second, to develop an algorithm and an associated interactive tool for creating predominantly hexahedral block structures.
Exploration and optimization of RAID architectures and virtualization technologies for high-performance data servers
Given the ever-increasing demands of numerical simulation, supercomputers
must constantly evolve to improve their performance and thus maintain a
high quality of service for users. These demands are reflected on storage
systems, which, to be performant, reliable, and capacitive, must contain
cutting-edge technologies concerning the optimization of data placement
and the scheduling of I/O accesses. The objective of this thesis is to
study these technologies such as GPU-based RAID and I/O virtualization,
to evaluate them, and to establish optimizations that can improve the
performance of HPC storage systems.
IO access scheduling on magnetic tapes using machine learning
Numerical simulations are used to obtain responses to physical phenomena that
cannot be reproduced, either because they are too dangerous or too expensive.
The models used for these simulations are increasingly complex, in terms of
size and precision, and require access to increasingly large computing and
data storage capacities. To this end, and in order to optimize costs, the use
of mass storage technologies such as magnetic tapes is critical. However, to
ensure good overall system performance, the development of algorithms and
mechanisms related to data placement and tape access scheduling is essential.
The objective of the thesis is to study the technology of magnetic tapes, as
well as existing mechanisms such as RAO (Recommended Access Order) or request
retention; and to implement new strategies for the optimization of magnetic
tape performance.