SIRIUS: Science-driven Data Management for Multi-tiered Storage   

This project explores the use of application level knowledge to optimize the
times to insight across a workload of multiple applications in a multi-user
environment with shared storage and network resources. The basis for this work
is the notion of selectable data quality to explore the tradeoffs between
accuracy of results, resource requirements, and time to insight on systems with
shared, oversubscribed computational and storage resources.   

Our thesis is that by adding application level knowledge about data to guide
storage system behaviors, we will obtain substantial benefits in the
organization, storage, and access to extreme scale data resulting in improved
productivity for computational science. We will demonstrate novel techniques to
facilitate efficient data placement onto multiple storage tiers, and enable
application-guided data reductions to address capacity, bandwidth, and latency.
Our goal is to address the associated data management challenges in the context
of current and emerging storage landscapes and expedite insights into mission
critical scientific processes. We will test our prototypes on current and
future Department of Energy (DOE) system with many of today's cutting edge
applications to ensure our techniques can be used within our framework on
current and next generation systems.