The assessment of the possible health effects linked to the exposure of humans to electromagnetic fields emitted from wireless systems is of particular interest to the well-being in our modern society.
Computer simulation is increasingly used for numerical dosimetry studies - meaning for the calculation and assessment of the radiation dose received by the human body. Usually these consist in solving a system of Maxwell equations, which model the propagation of electromagnetic fields in living tissues. GERShWIN (discontinuous GalERkin Solver for microWave INteraction with biological tissues) is a simulation software that has been developed for this purpose. It relies on a high order DGTD (Discontinuous Galerkin Time-Domain) finite element type method, which is designed to deal with fully unstructured tetrahedral meshes defining geometrical models of human tissues.
From an algorithmic point of view, the GERShWIN software combines sparse linear algebra operations at the global level (i.e. at the global mesh level) with dense linear algebra operations at the local level (i.e. at the element level). The software consists of three phases: a pre-processing phase, the time-stepping loop (the core of the DGTD method) and a post-processing phase. Because of its mixed sparse/dense structure, the DGTD method is perfectly suited to the DEEP/-ER architecture. The pre- and post-processing phases can be distributed to the Cluster Nodes, while the time-stepping phase can be offloaded completely on the Booster Nodes.
Developments within the DEEP-ER project
- In a first step, a hybrid MPI/OmpSs parallelisation strategy of the time-stepping phase is designed, implemented and evaluated for taking full benefits of the processing capabilities of the DEEP-ER Booster Nodes. The pre- and post-processing phases are parallelised using a classical approach combining partitioning of the tetrahedral mesh and MPI-based parallel programming.
- A second objective is to exploit the advanced I/O features of the DEEP-ER prototype in the context of multi-parametric studies (calculation and visualisation of the electromagnetic field distribution in the tissues for different configurations of problem parameters such as the position of the emitting source and the electromagnetic characteristic of the tissues, i.e. electric permittivity and electric conductivity). For that purpose, specific modules are developed based on the functionalities that are made available through the SIONlib library.
- A third objective is to design and implement a fault-tolerance strategy in the main time-stepping loop of the GERShWIN software in order to exploit the resiliency hardware/software infrastructure of the DEEP-ER prototype, in particular for checkpointing and restarting a simulation due to a failure at a Booster Node.
