Abstract:
External storage systems, such as parallel file systems, provide a number of important capabilities to today's HPC systems: holding checkpoints that allow applications to tolerate failures, temporarily storing data for subsequent analysis, and preserving science results. However, scaling the external storage model to provide the necessary I/O bandwidth and capacity is straining power, cooling, and cost budgets. The development of inexpensive, nonvolatile memory technologies such as flash memory and phase change memory makes it feasible to include solid-state, persistent memory inside future extreme-scale systems. The goal of the NoLoSS project is to conduct a detailed assessment of the potential roles and benefits of in-system storage in extreme-scale computational science, so that we can build more effective future systems for scientific discovery. |
