P6: Scalable Computing Techniques
This project will leverage existing technologies developed in computer science to realize and optimize the computational tasks involved in the computational workflows in the projects from P1 through P4. As the first tasks, the productive workflow environments, SimStack and FireWorks, used currently in two of the groups, will be disseminated to all application projects from P1 through P4. This will include deployment of the middleware on high performance computing resources (such as the ForHLR at KIT), support for the graduate students to integrate their methods in one of these environments, as well as teaching measures such hands-on tutorials and regular seminars.
P6 will have a strong research focus on the development of novel better scaling workflow methods in cases where the existing workflow frameworks have shortcomings. This activity will include usability and efficiency evaluation of existing workflow frameworks allowing tight coupling, in particular for QM/MM applications from P4 and P5, as well as available multiscale modelling languages and workflow programming languages. The thus selected framework will be extended with new components to enable the uptake in materials science community and further adapted to satisfy the domain-specific requirements of the tightly coupled models of projects P4 and P5. An efficient solution for tightly coupled problems will allow integration of asynchronous tasks taking into account dependencies and concurrency, and ensuring sufficiently short latency times and high data rates across the links between subsequent tasks. A combined solution exploiting the message-passing and dataflow programming paradigms will be pursued, thus achieving a high execution performance and improved scalability.