Layers of Abstraction and the Organization of Repair in High Performance Computing

• In Event: Infrastructures, Maintenance and Care - I

Abstract

High Performance Computing (HPC) is increasingly important scientific infrastructure. As in many areas of computing, designers and users of HPC systems rely on layers of abstraction to systematically manage tensions between hiding and exposing infrastructural complexity. Based on a study of the HPC community at a U.S. national laboratory, this paper explores how layers of abstraction are used to manage complexity and organize repair work in HPC, particularly when new computing hardware disrupts abstractions crucial to code developers. In this context, layers of abstraction function not as sharp boundaries between levels of infrastructure, but rather as coordinating interfaces that facilitate articulation, alignment, and repair work. This study suggests that the organization of infrastructure systems may structure opportunities for maintenance and repair, whether by accident or design, in ways that both provide for and constrain possibilities for systemic adaptation and change.

Author

• Benjamin Hayden Sims, Los Alamos National Laboratory