e-Seminar: "Biomedical supercomputing applications"
20 October 2021
The CompBioMed Centre of Excellence will host a two-hour e-Seminar entitled "Biomedical supercomputing applications" at 2pm CET on November 9, 2021. Presented by Dr Mariano Vazquez and Dr Guillaume Houzeaux (both Barcelona Supercomputing Center), the talk will focus on the Alya application, a simulation code for HPC mechanics, which is rightly famous for scaling to huge numbers of cores.
The first hour will cover Alya's uses in the biomedical field including time-critical, actionable cardiovascular simulations, whilst the second hour will reveal exactly how Alya achieves such excellent scaling.
This will be a highlight of the CompBioMed e-Seminar series for those interested in scalability for the exascale era.
For more information and to register, please visit www.compbiomed.eu/compbiomed-e-seminar-19
Abstract
Organ-level simulations represent both a challenge and an opportunity: the Virtual Patient modelling for precision medicine. It is the ultimate example of supercomputers usage for multiphysics/multiscale modelling problems. At organ level, modelling is done by tightly coupling different physics (eg fluid, tissue, electrophysiology, chemical reactions, heat, transport of large bodies, particles or species) with contributions from different time and space scales (cells, tissue, organ, system). To make things more complex, these problems present issues such as patient variability and comorbidities in complex geometries, with an extremely difficult validation. A way of facing these two issues is to run problems in a virtual patients population. All these things together make the use of supercomputers a decisive factor. In this talk we will define the challenges, and showcase them with some examples of the cardiovascular and respiratory systems.
To make those multiphysics simulations possible and efficient, several algorithmic and computational optimizations have been carried out. In particular, we will present classical actuations to enhance the strong and weak scalabilities of the code. These scalabilities are necessary to enable the road to exascale, but they should always be accompanied by efficiency measures. To this end, we have developed several original strategies that will be presented: a dynamic load balance strategy at node level; a co-execution strategy for heterogeneous architectures; a computationally enhanced Gauss-Seidel method for multiphysics coupling; an auto-tuning strategy to select the resources under efficiency criteria.