EAGER: Improving Lifetime and Data Reconstruction Time of Flash-Based RAID by Orchestrating Control between RAID and SSD Controllers

为了满足大数据应用的严格延迟和吞吐量要求，基于闪存的固态驱动器阵列（SSA）在企业存储系统中变得流行。与传统硬盘驱动器（HDD）相比，SSA具有许多优点，例如增强的输入/输出（IO）性能和存储密度，更低的功率、冷却和机架空间成本。然而，SSA的耐用性和寿命比HDD低，需要更频繁的更换，从而抵消了它们的一些优点。为了优化SSA的寿命，本项目将数据分为两个单独的类别，以下称为普通数据和近似数据。使用近似存储，包含具有不可纠正错误的页的坏块可以被重新用于存储可以容忍某些错误的近似数据。在SSA中应用近似存储需要新的机制和技术来协调廉价磁盘冗余阵列（RAID）和SSD控制器来管理近似数据和普通数据。本项目将（1）开发新的机制来协调RAID和SSD控制器之间的控制，以延长SSA的使用寿命;（2）研究新的方案，利用SSD的计算能力来加速SSA在繁重工作负载下的数据重建;以及（3）整合上述两种方法，以开发一个实验框架进行评估。预计该项目将对工业产生重大影响，因为它解决了一个真实的、非常及时的技术问题。此外，该项目在降低数据中心的存储成本、对研究生和本科生进行云计算架构的教育以及寻求扩大来自不同背景的学生的参与方面具有社会效益。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

To meet stringent latency and throughput requirements for big data applications, flash-based Solid-State Drive arrays (SSAs) have become popular in enterprise storage systems. SSAs have many advantages over conventional Hard Disk Drives (HDD), such as enhanced input/output (IO) performance and storage density, lower power, cooling, and rack space cost. However, SSAs have lower endurance and lower lifetime than HDDs, requiring more frequent replacement, negating some of their advantages.To optimize the lifespan of SSAs, this project will divide data into two separate classes, hereafter described as ordinary data and approximate data. With approximate storage, bad blocks that contain pages with uncorrectable errors can be reused to store approximate data that can tolerate some errors. Applying approximate storage in SSAs requires new mechanisms and techniques to coordinate redundant arrays of inexpensive disks (RAID) and SSD controllers to manage both approximate and ordinary data. This project will (1) develop new mechanisms to orchestrate control between RAID and SSD controllers to extend the lifespan of SSAs; (2) investigate new schemes to utilize the computing power of SSDs to accelerate the data reconstruction of SSAs under heavy workloads; and (3) integrate the above two approaches to develop an experimental framework for evaluation. This project is expected to have a major impact on industry, since it addresses a real, very timely technical problem. Further, this project has societal benefits, in reducing storage cost at data centers, educating graduate and undergraduate students about cloud computing architecture, while seeking to broaden participation of students from diverse backgrounds.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.