CAREER: Towards Learning-Based Storage Systems with Hardware-Software Co-Design

如今的存储系统已经构建成一个复杂的生态系统，其中涉及到存储设备、存储软件和应用程序级数据存储的开发和部署。为了快速满足不断增长的存储性能和效率要求，整个存储硬件和软件堆栈需要以协调的方式即时适应。然而，用目前的人为驱动的系统建设方法来实现这一点是具有挑战性的。机器学习技术的最新进展表明，基于学习的方法是解决系统优化问题的一种很有前途的方法。但是，目前尚不清楚应如何推进存储生态系统以采用学习技术来促进其在整个堆栈中的开发、部署和优化。该项目建议开发系统和体系结构技术，以构建基于学习的存储生态系统。它有四个主要推力。首先，该项目建议通过自动调整硬件规格来支持针对特定应用类型的定制存储设备的开发，因此，我们可以让开发人员以更少的时间和精力确定最佳设备规格。其次，该项目计划使用强化学习为多租户应用开发弹性存储管理，从而实现提高资源利用率和性能隔离。第三，项目提出将存储硬件知识融入到学习过程中，促进基于学习的存储软件的开发。最后，该项目将重新审视用于构建基于学习的存储驱动器的存储硬件体系结构，以进一步增强基于学习的存储生态系统。该项目将促进以内存和存储技术为中心的新课程的开发。该项目还将通过各种研究研讨会和夏令营向代表不足的高中生推广计算机系统和体系结构教育。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Storage systems today have been built into a complicated ecosystem, which involves the development and deployment of storage devices, storage software, and application-level data stores. To rapidly meet the ever-increasing storage performance and efficiency requirements, the entire storage hardware and software stack need to adapt instantly in a coordinated fashion. However, it is challenging to achieve this with current human-driven systems-building approaches. Recent advancements in machine learning techniques show that the learning-based approach is a promising method to solve system optimization problems. However, it remains unclear how the storage ecosystem should be advanced to embrace the learning techniques to facilitate its development, deployment, and optimizations across the full stack. This project proposes to develop systems and architecture techniques to build a learning-based storage ecosystem. It has four major thrusts. First, the project proposes to enable the development of customized storage devices for specific application types with automated tuning of hardware specifications, therefore, we can enable developers to identify optimal device specifications with much less time and effort. Second, the project plans to develop elastic storage management for multi-tenant applications using reinforcement learning, thus, we can achieve both improved resource utilization and performance isolation. Third, the project proposes to integrate the storage hardware knowledge into the learning procedure to facilitate the development of learning-based storage software. Finally, the project will revisit the storage hardware architecture for building learning-based storage drives to further enhance the learning-based storage ecosystem. The project will facilitate the development of a new course that centers around memory and storage technologies. The project will also promote computer systems and architecture education to underrepresented and high school students through various research workshops and summer camps.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.