SCoRE4HPC: Self-Configuring Runtime Environment for HPC Applications
Kevin S. Griffin · OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) · 2018
We introduce a framework called SCoRE (Self-configuring Runtime Environment) for optimizing performance of increasingly complex computer systems.Current petascale and future exascale platforms provide High-Performance Computing (HPC) applications with a set of distributed, diverse devices with increased memory capacity, advanced memory hierarchies, improved I/O capacity and bandwidth, and high levels of concurrency resulting in an exponential increase in overall system peak performance.To take advantage of these complex systems, users and programmers must exhaustively learn how to tune their applications for a specific hardware configuration.Tuning must be done for each optimization strategy (e.g., performance, throughput, power consumption) and for each hardware configuration.With the number of implementation possibilities becoming too large for manual optimization, there is a need for automatic approaches.We present SCoRE's framework and discuss the capabilities of its first implemented component, the supervised machine learning component (SMLC), using a case study of an HPC proxy application called Kripke.