HPC Needs a Tool Strategy (2005)

Authors: Michael L. Van De Vanter, Douglass E. Post (Los Alamos National Laboratory) and Mary E. Zosel (Lawrence Livermore National Laboratory)

The High Productivity Computing Systems (HPCS) program seeks a tenfold productivity increase in High Performance Computing (HPC). A change of this magnitude in software development and maintenance demands a transformation similar to other great leaps in industrial productivity. By analogy, this requires a dramatic change to the "infrastructure" and to the way software developers use it. Software tools such as compilers, libraries, debuggers and analyzers constitute an essential part of the HPC infrastructure, without which codes cannot be efficiently developed nor production runs accomplished.

The underappreciated "HPC software infrastructure" is not up to the task and is becoming less so in the face of increasing scale, complexity, and mission importance. Infrastructure dependencies are seen as significant risks to success, and significant productivity gains remain unrealized. Support models for this infrastructure are not aligned with its strategic value.

To achieve the potential of the software infrastructure, both for stability and for productivity breakthroughs, a dedicated, long-term, client-focused support structure must be established. Goals for tools in the infrastructure would include ubiquity, portability, and longevity commensurate with the projects they support, typically decades. The strategic value of such an infrastructure necessarily transcends individual projects, laboratories, and organizations.

Second International Workshop on Software Engineering for High Performance Computing System Applications , St. Louis, Missouri, May 15, 2005

5 pages (PDF)