CNS Core: Medium: Reimagining the Storage Stack for Emerging Memory Technologies

University Of Rochester, Rochester NY

Investigators

Abstract

Technological innovations are leading to increasing numbers and varieties of data storage devices. Important application domains are simultaneously becoming more data centric, and many of the most significant design challenges revolve around access to large, richly structured sources of information. To help programmers manage the complexity of data and devices, this project is developing new ways to organize and access data, to automate its movement and placement, and to protect it from accidental loss or corruption. By leveraging programmer specifications of application behavior and of performance, protection, and fault tolerance objectives (and by inferring these where possible), developed techniques will efficiently and automatically move data among devices including 3D-stacked and new nonvolatile alternatives with differing bandwidth, latency, and persistence properties. Fast, convenient access to byte-addressable nonvolatile memory will be provided through a new persistent segment abstraction. Targeted platforms will range from individual mobile or desktop machines to rack-level systems with "disaggregated" memory blades. As computing moves into the era of big data, techniques to manage large, fast, persistent memories have the potential to facilitate breakthroughs across the whole range of human endeavor, in government, industry, science, the arts, and entertainment. Students will gain experience with memory technologies and management techniques through new instructional modules in operating systems, programming languages, and parallel and distributed computing courses at the University of Rochester. Efforts to increase participation among traditionally underrepresented groups, in both the laboratory and the classroom, will complement existing department activities, including participation in the ABI/HMC BRAID initiative. Software and data developed under the project will be made publicly available under standard open-source licenses using platforms such as GitHub, and will be accessible for at least three years past the end of the grant. They will be shared with industrial contacts at Google, Facebook, IBM, Intel, AMD, Oracle, and others. Publications associated with the project will be made available at http://www.cs.rochester.edu/u/sandhya/publications.html and at http://www.cs.rochester.edu/u/scott/papers. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.