SHF: Small: Characterizing and Optimizing 3D NAND Flash

点击翻译按钮获取中文摘要

Inexpensive, reliable and large storage capacities are desirable in almost all application domains of national importance. For example, applications such as climate prediction, nuclear simulation and computational chemistry manipulate enormous datasets. A specific type of storage products, called NAND Flash, are suitable candidates for storing such large datasets given their performance, reliability and power efficiency advantages over traditional disk drives, and thus have seen widespread adoption recently. However, with continuous increase in dataset sizes, conventional NAND products available in the market are experiencing difficulty in coping with high performance and high reliability demands. 3D NAND flash is the newest architectural enhancement in the flash world to address this scalability problem, by enabling flash-memory cells to be stacked vertically in multiple layers. Despite the promise of 3D NAND flash, its fundamental characteristics and internal structures are not well understood as of today, from both the workload behavior and resource utilization perspectives. Yet architectural, firmware and system software level problems and potential opportunities have largely been ignored, which this project seeks to address. The outcome of this project is expected to enhance the design and utilization of 3D NAND storage in many big data application domains. The educational and outreach components of this project includes (1) a new graduate level course on advanced storage system concepts; (2) engaging undergraduates from Penn State's Schreyer Honors College in the research activities; (3) involving students from the Integrated Undergraduate/Graduate (IUG) program at Penn State to interest them in this topic, for possible graduate study in this project; and (4) organizing summer camps for high school girls and high school teachers.Specifically, this project conducts an in-depth study of 3D NAND flash memory (in particular solid state devices built using them), with the intent of identifying the workload and resource utilization characteristics of 3D NAND flash, implementing and evaluating a set of architectural, firmware and system software level optimizations targeting 3D NAND flash architectures as well as 2D/3D NAND flash hybrids. The project consists of four tasks: (1) performing a thorough exploration of various architectural optimizations specific to 3D NAND flash; (2) investigating firmware-level optimizations such as long read latency mitigation, as well as various reliability and lifetime enhancement techniques; (3) conducting an in-depth study of system software level issues such as file system-level request scheduling; and (4) performing a detailed experimental evaluation of the optimizations using different types of application workloads and an in-house 3D NAND simulation infrastructure.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.