This postdoc position is announced within the collaboration between Inria Lyon (ROMA team) and Simula (High Performance Computing Department) Norway.
Location
Inria Lyon, ROMA team. The ROMA team aims at designing models, algorithms, and scheduling strategies to optimize the execution of scientific applications on High-Performance Computing platforms.
Research topic
Real-world applications of finite element mesh (FEM) simulations require parallel computing. In such a setting, computational meshes must be partitioned among multiple processes as a first step. One important goal of mesh partitioning is to keep the resulting inter-process communication overhead low while achieving load balance for scalability.
The current state-of-the-art and mainstream approaches treat mesh partitioning as a graph or hypergraph partitioning problem, which is then tackled by using well-known libraries. The common goal of these libraries is to minimize the total volume of inter-process communication; some additional objectives such as the total number of communication operations are sometimes addressed in an ad hoc manner. The aim of this post-doc position is to define accurate communication-cost objective functions for parallelizing FEM simulations in the current HPC systems while taking the topology of the targeted systems into account, and then to develop a mesh partitioner for reducing the communication cost. Generality in the partitioner will be achieved by using graph models, and developing algorithms on those models.
