Interactive On-Demand HPC at Lincoln Laboratory and Beyond

Abstract

Since their emergence in the 1980s, supercomputers have been considered high-value assets, and like telescopes, they have been scheduled with batch queue and reservation systems. However, human-in-the-loop research, rapid prototyping, and engineering activities do not lend themselves well to such scheduling. For over 20 years, MIT Lincoln Laboratory Supercomputing Center and several other organizations have been researching, developing, and implementing an alternative paradigm: interactive, on-demand HPC. In this talk, we will overview a number of facets of implementing interactive, on-demand HPC capabilities including scheduling and organizational policies, software tools, system design, data management, user support, and user education/training.