Collaborative Research: SHF: Medium: A New Direction of Research and Development to Fulfill the Promise of Computational Storage

随着处理器和存储之间的速度差距不断扩大，使计算更接近数据是高速数据处理的一个有希望的方向。然而，许多具有挑战性的问题，如设备硬件上的有限资源、严格的成本和功耗限制、软件开发的困难以及缺乏系统的解决方案，不幸地阻碍了计算存储在现实中的广泛采用。本项目通过使用内聚方法将计算存储转变为整个系统中的协作组件来解决这些关键问题。本项目旨在通过研究活动培养不同层次的学生，以新的研究成果丰富课程和课堂教学，并为教育和外展活动做出贡献，从而产生更广泛的影响。本项目致力于解决具有挑战性的研究问题，旨在为实现计算存储的承诺提供新的方向。通过使用整体和面向系统的方法，该项目在系统层次的多个层面调查关键研究问题，并制定有效的解决方案。具体而言，该项目研究了多个重要方面，以便系统地将计算存储集成到现有的计算生态系统中，例如为应用设计面向服务的抽象，优化系统级资源利用，利用邻近性和减轻设备硬件中的内存资源争用，调整应用的核心数据结构和算法以充分利用异构计算资源等。为了有效地利用计算存储，还研究了一组具有代表性的应用案例。该项目的成功将为使计算存储能够应对日益以数据为中心的应用中的关键挑战做出广泛而重要的贡献。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

As the speed gap between processor and storage keeps widening, moving computation closer to data is a promising direction for high-speed data processing. However, multiple challenging issues, such as the limited resources on device hardware, stringent cost and power constraints, difficulties in software development, and the lack of systematic solutions, are unfortunately hindering a widespread adoption of computational storage in reality. This project addresses these critical issues by using a cohesive approach to turn computational storage into a cooperative component in the whole system. This project also aims to make a broader impact by training students at different levels with research activities, enriching curriculum and classroom teaching with new research results, and contributing to educational and outreach activities.This project makes an effort to address the challenging research issues, aiming to provide a new direction to fulfill the promise of computational storage. By using a holistic and system-oriented methodology, the project investigates critical research issues across multiple layers in the system hierarchy and develops effective solutions. Specifically, the project studies multiple important aspects in order to systematically integrate computational storage into existing computing ecosystems, such as designing a service-oriented abstraction for applications, optimizing system-level resource utilization, leveraging proximity and mitigating memory resource contention in device hardware, adapting core data structures and algorithms of applications to fully exploit heterogeneous computing resources, etc. A set of representative application cases is also studied for effectively leveraging computational storage. The success of this project will make broad and significant contributions to enable computational storage to address critical challenges in increasingly more data-centric applications.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.