Date of Award
5-11-2025
Date Published
June 2025
Degree Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Electrical Engineering and Computer Science
Advisor(s)
Bryan Kim
Keywords
Flash memory;SSD;Storage system
Subject Categories
Computer Sciences | Physical Sciences and Mathematics
Abstract
The explosive growth of data has led to increased attention on NAND flash-based solid-state drives (SSDs), which offer high performance, low power consumption, and significant capacity per unit volume when compared to traditional hard-disk drives (HDDs). However, as NAND flash memory density continues to scale, modern SSDs suffer from what is known as fail-slow symptoms, and their performance degrades over time as they wear out. In this dissertation, we focus on understanding and addressing the performance, reliability, and sustainability challenges of modern flash-based storage systems by modeling key metrics, analyzing the design tradeoffs between these metrics, and optimizing existing storage I/O stack across various layers. The first part of the dissertation presents our comprehensive study on SSD aging and fail-slow symptoms. Flash memory has a limited lifetime and suffers reduced reliability as the number of program/erase (P/E) cycles increases. To understand the aged behavior of SSDs, we first propose fast-forwardable SSD (FF-SSD), a machine learning-based SSD aging framework that generates representative future wear-out states without enduring prohibitive simulation times. FF-SSD incrementally builds lightweight regression models for flash blocks to capture the changes in SSD-internal states and predicts their trajectory using the information from past executions. With that, we further conduct an in-depth study on existing SSD lifetime extension algorithms and uncover that conventional wear leveling methods are far from perfect. Our finding challenges the established norms of SSD design and calls for a reevaluation of wear leveling’s role. The second part of the dissertation introduces the design and implementation of a capacity-variant storage system (CVSS) for modern SSDs. In the current capacity-invariant interface where the storage device exports a fixed capacity throughout its lifetime, an SSD has no other choice but to trade performance for reliability as it wears out. Instead, CVSS redesigns the existing storage interface from a holistic perspective by exploiting tradeoffs among capacity, performance, and reliability (CPR) and proposes a flexible capacity-variant interface that allows the SSD to maintain performance by gracefully reducing the number of exported blocks. The third part of the dissertation proposes to improve the performance and sustainability of all-flash array (AFA) storage under heterogeneous device environments. The overall architecture of existing AFA systems is built around the assumption that the underlying storage components are homogeneous. This architecture results in significant disk under-utilization when considering heterogeneity among storage devices, particularly for modern SSDs. In response to these challenges, we present paRAID (placement asymmetric RAID), a heterogeneity-aware AFA system that transitions from traditional symmetric architectures to an asymmetric architecture, enabling the full utilization of all storage devices for optimized performance and sustainability.
Access
Open Access
Recommended Citation
Jiao, Ziyang, "TOWARDS THE NEXT GENERATION OF STORAGE STACK FOR NAND FLASH MEMORY-BASED SYSTEMS" (2025). Dissertations - ALL. 2092.
https://surface.syr.edu/etd/2092