Project Description

Understanding the ORNL and DOE HPC application landscape is vitally important for the integration of heterogeneous computational architectures and exploration of future computing technologies. Currently, there are simple questions about our applications that we cannot answer quickly and accurately, like which programming languages, parallelization features, libraries, and communication APIs are most important for supported applications. These questions become even more urgent for the documentation of application requirements for next generation HPC systems, the planning of long-term computer science research programs to fill capability gaps, and in the execution of scientific applications readiness programs such as ORNL’s Center for Accelerated Applications Readiness (CAAR), which is preparing codes to accomplish large-scale science on Summit. To provide this critical yet currently unavailable information, we need to automate the collection of application program characteristics from a variety of tools, such as compilers, debuggers, performance tools, and cluster management/scheduling systems. We will create a database to provide convenient access to this information, enabling data analytics and knowledge discovery techniques to inform ongoing HPC research and strategic planning at our facilities. These cross-application analyses will lead to an understanding of the overall HPC application landscape and where we need to invest in system software and tools. Participants include:
Jack C. Wells (PI) (NCCS)
Oscar Hernandez (CSMD)
Matthew Graham Lopez (CSMD)
Reuben D. Budiardja (NCCS)