Anshu Dubey

Multi-scale, multi-physics simulation codes take years to develop, with a typical life cycle that can span several generations of high performance computing (HPC) platforms. For these codes effective utilization of HPC resources is a balancing act between portability and performance. Ongoing deep hardware architecture changes have brought this balancing act to a point where codes cannot cope without more programming abstractions. Under National Science Foundation’s SI2 program, several code teams joined together to conceptualize a software institute that could facilitate the exploration of possible solutions. FLASH, a highly capable code which serves astrophysics, HEDP and several other scientific communities, was one of the participants.In this presentation I will describe FLASH in the context of software engineering evolution, current challenges, and conceptualization for the future. I will also describe FLASH’s current architecture and present a design proposal that incorporates some of the more promising programming abstractions.