DPW-LRU: An Efficient Buffer Management Policy Based on Dynamic Page Weight for Flash Memory in Cyber-Physical Systems

Youwei Yuan, Jintao Zhang, Guangjie Han, Gangyong Jia, Lamei Yan, Wanqing LiIEEE Access 2019 · DOI

Abstract

Owing to its high performance, small size, and low energy consumption, NAND flash memory has been extensively adopted in cyber-physical systems. However, the inherent characteristics of flash memory, including not-in-place update and asymmetric I/O latencies, present difficulties in the design of buffer management policies. In this paper, we propose an enhanced buffer management policy for the flash memory referred to as dynamic page weight least recently used (DPW-LRU), which considers temporal locality and simultaneously provides effective utilization of limited buffer resources. Page migration is further enhanced by identifying the page access mode and frequency while separating the buffer into two different regions. A novel eviction algorithm is also designed to reduce the write operations and maintain a high hit ratio of the buffer regions, combining dynamic temporal locality, real-time eviction cost, and recency of pages. The experimental results show that DPW-LRU improves the hit ratio by up to 8.3%, decreases the write operation by up to 22.6%, and reduces the overall latency by up to 18.8% relative to those of other state-of-the-art buffers management policies.

Introduction

In Cyber-Physical Systems (CPS)—such as autonomous vehicles and industrial IoT—data throughput and latency are critical performance metrics. NAND flash memory has largely replaced traditional hard disk drives (HDDs) in these environments due to its superior speed and compact form factor. However, flash memory introduces unique architectural challenges that standard software layers often fail to address.

System architecture of flash-based storage in cyber-physical systems.

Unlike HDDs, flash memory is characterized by asymmetric I/O latencies: read operations are significantly faster than write and erase operations. Furthermore, flash memory requires "out-of-place" updates, where an entire block must be erased before a page can be overwritten. Standard buffer management policies like Least Recently Used (LRU) treat all data pages equally, often leading to the premature eviction of "dirty" (modified) pages. This triggers expensive write operations, degrading both system performance and the physical lifespan of the flash hardware.

The DPW-LRU Methodology

To address these issues, we propose DPW-LRU (Dynamic Page Weight Least Recently Used). This policy optimizes buffer management through a dual-region architecture and a sophisticated eviction algorithm based on hardware-aware weighting.

Dual-Region Architecture

The buffer is partitioned into two distinct areas:

  1. Working Region: Dedicated to "hot" pages that are accessed frequently.
  2. Exchange Region: Serves as a landing zone for "cold" pages or those recently evicted from the working region.

This separation effectively prevents "buffer pollution," where a sudden burst of sequential read requests might otherwise flush out frequently used data.

Architecture of the DPW-LRU buffer management policy.

Dynamic Page Weight (DPW) Calculation

The core of the eviction logic is the Dynamic Page Weight ($P^w_i$), calculated for each page based on three primary factors:

  • Temporal Locality Interval (TLI): Forecasts future access based on historical intervals.
  • Eviction Cost (EC): Incorporates real-time hardware latency. Dirty pages are assigned higher weights to reflect the higher cost of writing them back to flash.
  • Recency (RE): Tracks the time since the last access to maintain data freshness.

When the buffer reaches capacity, the system evicts the page with the lowest weight, ensuring that the most valuable or "expensive" data remains in the fast RAM buffer.

Experimental Evaluation

The DPW-LRU policy was evaluated using the Flash-DBSim platform across various workloads, including synthetic Zipfian traces and real-world datasets like Financial1 (OLTP) and WebSearch1.

Compared to state-of-the-art policies such as CF-LRU and AD-LRU, DPW-LRU demonstrated significant improvements:

  • Hit Ratio: Increased by up to 8.3%, reducing the frequency of slow flash accesses.
  • Write Reduction: Decreased flash write operations by up to 22.6%, extending hardware longevity.
  • Latency: Reduced overall I/O latency by up to 18.8%, meeting the real-time demands of CPS.

Conclusion

DPW-LRU offers a hardware-aware approach to buffer management that significantly improves performance and durability for flash-based storage in cyber-physical systems. By integrating real-time eviction costs and a dual-region structure, it effectively balances hit ratios with write minimization. Future work will focus on integrating this policy into database management systems and exploring its performance on open-channel SSDs.

Summary updated

Abstract

Owing to its high performance, small size, and low energy consumption, NAND flash memory has been extensively adopted in cyber-physical systems. However, the inherent characteristics of flash memory, including not-in-place update and asymmetric I/O latencies, present difficulties in the design of buffer management policies. In this paper, we propose an enhanced buffer management policy for the flash memory referred to as dynamic page weight least recently used (DPW-LRU), which considers temporal locality and simultaneously provides effective utilization of limited buffer resources. Page migration is further enhanced by identifying the page access mode and frequency while separating the buffer into two different regions. A novel eviction algorithm is also designed to reduce the write operations and maintain a high hit ratio of the buffer regions, combining dynamic temporal locality, real-time eviction cost, and recency of pages. The experimental results show that DPW-LRU improves the hit ratio by up to 8.3%, decreases the write operation by up to 22.6%, and reduces the overall latency by up to 18.8% relative to those of other state-of-the-art buffers management policies.
BibTeX
@inproceedings{yuan2019dpw,
  author = {Youwei Yuan and Jintao Zhang and Guangjie Han and Gangyong Jia and Lamei Yan and Wanqing Li},
  title = {{DPW-LRU: An Efficient Buffer Management Policy Based on Dynamic Page Weight for Flash Memory in Cyber-Physical Systems}},
  booktitle = {IEEE Access},
  year = {2019},
  doi = {10.1109/access.2019.2914231},
}
Figure