Challenges in Advanced Computing - Multi-this and Multi-that

Hans-Joachim Bungartz Scientific Computing chair at TUM's informatics department Professional outline Abstract:Advanced Computing is nowadays somewhat established as describing the field of dealing with large-scale simulations, and doing this beyond the classical foci of mathematical models and numerical schemes. Actually, implementing algorithms in a hardware-efficient way, managing large simulation research codes, or exploring huge sets of data produced by the latter also turned out to be crucial when striving for "insight, not numbers". As a result, the challenges are more diverse than before, some of them being expressed by a "multi-X" notion (multi-disciplinary, multi-physics, multi-dimensional, multi-scale, multi-level, multi-core, ...), and they are highly interwoven.

The first part of the presentation will provide an overview of current challenges in Advanced Computing, in particular those which are not yet at a prominent place on the agenda. In the second part, two examples from our current research activities will be presented: our PDE framework Peano, where space-filling curves are used as the general design pattern, governing grid generation, adaptive refinement, data traversal, solver construction, and parallelization; and spatially adaptive sparse grids for high-dimensional problems such as classification.

Practical Methodologies of Co-Design for Exascale Systems and Applications

Adolfy Hoisie Director of the Center for Advanced Architectures at the Pacific Northwest National Laboratory Professional outline Abstract:For the coming decade the HPC community is embarking on an exciting path towards Exascale computing. With this growth in size, we are witnessing an exponential increase in the complexity of the systems and applications. In this presentation we will describe the tools-of-the trade for optimal co-design of systems and applications at Exascale. The process of capturing the workload represented by entire applications into accurate predictive models will be described.

We will walk through the process of predicting system performance for the application workload considered using these models. A particular emphasis will be placed on methodologies that are practical, accurate, and allow for the possibility of addressing the problem for full systems and applications. We will conclude by summarizing a number of factors that in our view will significantly impact the development and performance of Exascale systems.