DSN 2016 Tutorial: Resilience for Scientific Computing: From Theory to Practice

Envisioning the future Exascale platforms requires a deep understanding of the impact of faults and a realistic view of the needs in terms of reliability. Reliability has been highlighted as one of the major concerns and the IESP projects an increase in node performance and concurrency by one or two orders of magnitude, which translates, even under the most optimistic circumstances, in at least one order of magnitude direct decrease of the mean time to interruption of capability platforms. Because of this tendency, platform providers, software developers, and high-performance application users who target capability runs, cannot regard the occurrence of interruptions due to a failure as a rare dramatic event, but must consider faults as an inevitable component of the computing eco-system, and design and develop software components that have some form of fault-tolerance integrated at their core.