Computing Patterns for High Performance Multiscale Computing (ComPat)

Multiscale phenomena are ubiquitous and key to understanding the complexity of our world. Despite the significant scientific progress enabled through computer simulations over the last decades, the capabilities for accurately and reliably simulating multiscale physical processes is still quite limited. Hence scientists in this field cannot really make the most of current high performance computing ressources in the tera- and petascale range.

In the ComPat project, experts develop multiscale computing algorithms capable of producing high-fidelity scientific results that are scalable to exascale computing systems. The core idea is to develop generic and reusable frameworks that make it possible to implement customized algorithms and thus optimise load balancing, data handling, fault tolerance and energy consumption under generic exascale application scenarios.

Implementing nine grand challenge applications which are exascale-ready demonstrate the viability of this approach. The ultimate goal is to establish new standards for multiscale computing at exascale, and provision a robust and reliable software technology stack that empowers multiscale modellers to transform computer simulations into predictive science.

LRZ is heading the implementation of an experimental execution environment on European facilities available to the project. The environment is used to measure performance characteristics and develop models that can provide reliable performance predictions for emerging and future exascale architectures. Special emphasis is put on tools to optimize energy efficieny.

Events and Conferences

Project meeting at IPP in Garching, 10-14 October 2016
Project meeting at Brunel University in London, 13-17 March 2017
ComPat Workshop at ICCS 2017 conference in Zurich, 12-14 June 2017

Publications

Alowayyed, S., Groen, D., Coveney, P., Hoekstra, A.: Multiscale Computing in the Exascale Era. In: arXiv preprint arXiv:1612.02467. 2016.

Fast Facts