SHF: Small: Redesigning the System Architecture for Ultra-High Density Data Storage

点击翻译按钮获取中文摘要

Computer storage systems are advancing towards extremely high density. With emerging storage technologies, such as Quad-Level Cell (QLC) and 3D vertical NAND, built using advanced 10 nanometer manufacturing processes, one single server will be able to accommodate petabytes of data in the near future. Such an unprecedented storage density growth allows us to build highly consolidated server systems at low costs, but in the meantime, it also brings a variety of critical challenges in terms of scalability, reliability, and shareability. Simply transplanting the existing system stack onto vastly condensed storage would create a severely unbalanced system - although the storage is able to accommodate a huge amount of data, the unoptimized hardware and software components in the system stack are unable to efficiently and reliably manage and deliver the data. Thus, the growth of storage density should be regarded as a fundamental system change, which demands a full-scale re-assessment on the whole system stack. This project systematically studies these challenging research issues by using a cohesive approach to reconsider the storage system architecture design at different levels. This project also aims to increase its impact by training students in research activities, contributing to educational activities, and outreach to K-12 students.The project addresses the critical technical challenges by adopting a holistic, cross-layer, whole-system approach for optimizations across multiple layers in the computer system hierarchy. Specifically, the project investigates various important aspects in the storage architecture design to re-balance the system for highly condensed storage, such as reconsidering the mapping structure for I/O efficiency, leveraging the QLC flash storage and non-volatile memory for efficient data management, offloading computationally intensive storage operations to hardware accelerators. The project also studies the intermediate layers in the system stack to manage hardware and provide simplified, semantic-rich storage services. A set of key data-intensive applications is further studied for effectively leveraging the next-generation high-density storage.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.