Distributed Resilient Systems

archived

The computational workflows associated with modern science are becoming increasingly complex, often processing an astounding amount of data generated by geographically-distributed instruments. The data is analyzed using a variety of local and remote compute resources and integrated with simulations and artificial intelligence (AI)--enhanced models to both direct the ongoing experiments and inform scientific progress. The future continuation of this trend is outlined by the 2022 report on Envisioning Science in 2050 [1]. Pursuing innovative research directions in techniques for advanced middleware and operating and runtime systems is critical to address the unprecedented challenges in implementing future workflows. These research directions may involve, but are not limited to, coordinating work on: a) billions of threads of execution on a supercomputer; b) several geographically-separated supercomputers; c) advanced experimental systems which produce hundreds of petabytes of data each day; and/or d), billions of distributed sensors monitoring the climate or other systems of interest. Recognizing that in systems of this size and complexity sporadic failures of individual components are inevitable, scientific workflows and their supporting middleware and system software must be designed with resilience in mind.

High