| Description |
At present, a signi#12;cant amount of ongoing scienti#12;c research relies on computational models. High-performance computing (HPC) resources are often required to obtain results in a reasonable amount of time. However, as physical and practical limitations constrain the performance progression of computer CPUs, high-performance systems must scale laterally and leverage parallelism to increase potential performance. As a result, modern HPC ecosystems are enormously complex, and require novel methods for performance analysis. We explore using Performance Co-Pilot, a set of software for distributed performance metrics collection, and other technologies to design and build a data pipeline with the end goal of developing predictive analytics for cost-e#11;ective HPC centers. We make substantial progress towards this goal, although we had limited results in the realm of analysis. |
