CSSI Frameworks: Performance Engineering Scientific Applications with MVAPICH and TAU using Emerging Communication Primitives

Modern HPC platforms are using multiple CPU, GPUs and high-performance interconnects per node. Unfortunately, state-of-the-art production quality implementations of the popular Message Passing Interface (MPI) programming model do not have the appropriate support to deliver the best performance and scalability for applications (HPC and AI) on such systems.This project aims to investigate and develop the following innovations by co-designing MVAPICH2 and TAU libraries to scale driving science domains---including AWP-ODC and heFFTe: 1) Load-aware designs for MPI asynchronous communication, 2) Cross runtime coordination for MPI+X applications, 3) Partitioned point-to-point primitives, 4) Application-aware neighborhood collective communication, 5) Support for adaptive persistent collective communication, and 6) Coordinating communication kernels on GPUs. Integrated development and evaluation are carried out to ensure proper integration of proposed designs with the driving applications, and closely work with internal and external collaborators to facilitate wide deployment and adoption of the released software. A synergistic and comprehensive development plan, involving personnel from OSU, SDSC, UO, UTK, is carried out. The transformative impact of the proposed effort is to extract the performance and scalability of HPC applications in next-generation HPC architectures through intelligent performance engineering.