Toward maximizing outcome with the Post-K computer, the Application Development Team conducts research and development with related researchers in architecture, system software and algorithm fields. Its core missions are as follows;
Co-design based on Target applications
"Target applications" of Post-K computer have been chosen by the advisory panel of MEXT as the priority challenges of social and scientific problems. We execute performance analysis and optimization of target applications to reflect co-design of application and system in cooperation with the executing agencies of target applications and the colleague teams in this project. The optimization methods are recorded for future development of applications. We also develop new algorithms and implementation methods in collaboration with Co-design team.
Development of an application benchmark suite for system performance evaluation
We develop and maintain "mini-apps" that are designed to represent the same performance characteristics as full applications from major computational science domains. It is designed to be compact and easy to run, and is made available in source code format, which is often not the case with the full applications. We study them from various analysis aspects with performance evaluation on supercomputers including K-computer. Our aim is to establish a mini-app benchmark suite that is widely and internationally used for system performance evaluation. Collaborative research with universities will be conducted, aiming for establishment of system performance evaluation methodologies.
Research and development of application infrastructures
Our team and R&D department of R-CCS develop application infrastructures including the libraries and frameworks which are important elements in maximizing the outcome of the users of Post-K computer.
The team also leads the promotional activities to continue the research work done through the "Feasibility Study on Future HPC Infrastructures (Application Working Group)", extracting the social and scientific challenges to be solved by HPC during next 5-10 years.