Challenges in Extreme Engineering Algorithms

Abstract

Engineering simulations like aircraft, helicopter or safe building design on massively parallel hardware result in extreme challenges regarding algorithm and software framework development. Algorithms must be robust, scalable and allow for highly accurate computations, in many cases for the solution of time-dependent and non-linear engineering problems. Software frameworks must exploit hardware parallelism on several levels, must be tuneable for different, heterogeneous computing systems and must be easily to maintain. The latter is usually supported by a component-wise framework structure. Engineering simulations often require coupling of several application components. This is a particular challenge for the underlying parallel algorithms, for example regarding convergence and communication behavior as well as load balancing. This session presents highly parallel frameworks and algorithms for helicopter design as well as for the simulation of seismic waves and building vibrations. The latter framework exploits autotuning for performance optimization.