RII Track-4: NSF: Scalable MPI with Adaptive Compression for GPU-based Computing Systems

The overarching goal of this project is to deliver a performant and scalable Message Passing Interface (MPI) library, to accelerate scientific applications running on cutting-edge computing systems leveraging scientific data compression. In particular, the proposed library will identify and compose the best-fit compression pipeline for message passing in an adaptive fashion, such that the trade-off between computation and communication can be well balanced. Success of this project will reduce the time to scientific insights for a wide range of applications spanning across computational science and artificial intelligence, making a high and long-lasting impact to community. It will also expand the research capacity of the home institution while contributing to workforce development and broadening participation in the jurisdiction.