High-performance computing (HPC) systems are designed for pursuing extreme-scale parallelism and computational power to accelerate various computational-expensive applications in scientific computing. HPC software stack is deep and wide, mainly due to the complex computational requirements from various domain-specific algorithms and the diverse capacities provided by the underlying hardware architecture. For instance, domain-specific language or multi-layer runtime library is a common approach for developers to decouple the complex scientific algorithms from parallel programming (i.e., deep). To fully utilize different hardware components, different parallel programming models and runtime subsystems often coexist in an application software stack such as the popular MPI + “X” programming model (i.e., wide). Although abstraction allows developers and researchers to isolate innovations for different software components, experiences have indicated that co-design is the key strategy to achieve exascale computing and beyond. Co-design across the deep and wide HPC software stack is a very challenging task. However, the potential performance gain can be significant. In this talk, I will share the experiences and insights from several projects that explored the co-design approach across applications and parallel runtime subsystems.

The discussion following this talk will be moderated by Mary Hall.

Dr. Min Si is an Assistant Computer Scientist at Argonne National Laboratory working with the Programming Models and Runtime Systems group. Min’s research interests include communication runtime in high-performance computing and parallel programming models. Min was previously an Enrico Fermi Postdoctoral Scholar of Argonne National Laboratory. She received the Ph.D. degree in computer science from the University of Tokyo. Her doctoral dissertation focused on the optimization of high performance message passing communication on massively parallel multi-/many-core architectures. Min is the recipient of the 2018 IEEE-CS Technical Consortium on High Performance Computing (TCHPC) Early Career Researchers Award for Excellence in High Performance Computing, and won the Karsten Schwan Best Paper Award at HPDC 2018.