Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Crash Consistency in DRAM-NVM-Disk Hybrid Storage System

Guoyu Wang, Xilong Che, Haoyang Wei, Chenju Pei, Juncheng Hu

TL;DR

This work tackles crash-consistency in a DRAM-NVM-Disk heterogeneous storage stack, focusing on the challenges that arise from differing write sequences and granularities across devices. Using NVPC as a case study, it formalizes sync semantics and crash-recovery models, and then proposes a three-pronged approach to ensure recoverable correctness: per-page persistent version metadata in NVM (Algorithm 1) to resolve data version after crashes, recording write-back events on the NVM (Algorithm 2) to manage arbitrary-length sync writes, and explicit write-back timing marks (Algorithm 3) to handle the uncertainty of write-back duration. The core contributions are the formal problem delineation and the design of concrete, provably corrective mechanisms that preserve crash semantics while preserving performance of the fast path. The findings provide a practical design space for heterogeneous storage systems adopting multiple write sequences and granularities, with direct implications for reliability and performance in systems integrating NVMe/NVM with traditional disks.

Abstract

NVM is used as a new hierarchy in the storage system, due to its intermediate speed and capacity between DRAM, and its byte granularity. However, consistency problems emerge when we attempt to put DRAM, NVM, and disk together as an efficient whole. In this paper, we discuss the challenging consistency problems faced by heterogeneous storage systems, and propose our solution to the problems. The discussion is based on NVPC as a case study, but can be inspiring and adaptive to all similar heterogeneous storage systems.

Crash Consistency in DRAM-NVM-Disk Hybrid Storage System

TL;DR

This work tackles crash-consistency in a DRAM-NVM-Disk heterogeneous storage stack, focusing on the challenges that arise from differing write sequences and granularities across devices. Using NVPC as a case study, it formalizes sync semantics and crash-recovery models, and then proposes a three-pronged approach to ensure recoverable correctness: per-page persistent version metadata in NVM (Algorithm 1) to resolve data version after crashes, recording write-back events on the NVM (Algorithm 2) to manage arbitrary-length sync writes, and explicit write-back timing marks (Algorithm 3) to handle the uncertainty of write-back duration. The core contributions are the formal problem delineation and the design of concrete, provably corrective mechanisms that preserve crash semantics while preserving performance of the fast path. The findings provide a practical design space for heterogeneous storage systems adopting multiple write sequences and granularities, with direct implications for reliability and performance in systems integrating NVMe/NVM with traditional disks.

Abstract

NVM is used as a new hierarchy in the storage system, due to its intermediate speed and capacity between DRAM, and its byte granularity. However, consistency problems emerge when we attempt to put DRAM, NVM, and disk together as an efficient whole. In this paper, we discuss the challenging consistency problems faced by heterogeneous storage systems, and propose our solution to the problems. The discussion is based on NVPC as a case study, but can be inspiring and adaptive to all similar heterogeneous storage systems.
Paper Structure (7 sections, 5 theorems, 3 figures, 3 algorithms)

This paper contains 7 sections, 5 theorems, 3 figures, 3 algorithms.

Table of Contents

  1. Introduction
  2. Heterogeneous Storage Systems
  3. Consistency Problems and Solutions in the System
  4. Problems from Write Sequence Difference
  5. Problems from Write Granularity Difference
  6. Problems from Write-back Duration
  7. Conclusion

Key Result

Theorem 1

If $(write,sync,crash,read)$, then when $read$ is performed, the data on the storage should be no earlier than the $write$.

Figures (3)

  • Figure 1: Space-time diagram for problematic cases caused by variant write sequence.
  • Figure 2: Space-time diagram for problematic cases caused by variant write granularity.
  • Figure 3: Space-time diagram for problematic cases caused by an uncertain write-back time.

Theorems & Definitions (11)

  • Definition 1
  • Definition 2
  • Definition 3
  • Theorem 1
  • Theorem 2
  • Proposition 1
  • proof
  • Proposition 2
  • proof
  • Proposition 3
  • ...and 1 more