Development of a pervasive simulation toolkit for biomechanical models

The project is part of a larger group of projects in PN7 that aim to develop a simulation toolkit consisting of various bio-mechanical simulation modules that allows for interactive pervasive simulations. PN7-2 particularly contributes to the hardware-specific optimization of simulation components fulfilling different user-defined constraints, e.g., in terms of runtime or accuracy. This comprises the definition of metrics characterizing each simulation module in terms of complexity, accuracy, and other requirements, the implementation of different solvers for different models targeting a selection of different architectures (e.g., mobile phone, laptop, cluster), and the development of automated scheduling tools for heterogeneous distributed systems as well as user and sensor interaction interfaces. The project is different from classical numerical methods and high-performance computing who often target the optimization of a single efficiency aspect such as grid adaptivity, the discretization order or the parallelization strategy for an otherwise fixed simulation setup. For unconventional architectures such as accelerator-based platforms, most of the effort is invested into porting and parallelisation, without reconsidering the underlying numerics in an hardware-aware way. The core idea of our approach is to consider the simulation environment for upper arm muscle simulation as a combination of a simulation scenario for which various choices are available in terms of model, (non-)linear solvers, floating point accuracy, and domain partitioning and load balancing strategy (developed in cooperation with PN5-1) and a dynamically changing and heterogeneous environment (user requests and compute hardware). We build our work on data ranging from hardware characteristics such as latencies, bandwidth, or clock-rates over user requirements such as accuracy limits and time constraints, to simulation meta-data such as runtimes and error estimates.