Maximizing the Return on Architecture via Workload Management

HPC architectures are the cumulative effect of processors, memory and interconnects taken together with software that supports the development environment. Because HPC architectures are ‘fractal’ in nature, i.e., having similar structure at various scales, it is often challenging to maximize the Return on Architecture (RoA) in practice. By arbitrating consumer-provider economics subject to policies, workload-management technology offers an improved return for isolated systems, and a significant return on more-complex environments that may scale to clusters or grids. Through several illustrative, technical examples, this presentation explores RoA for the case of capability computing.