Lune

ISCA2026顶会

LOONG: Utilizing Long-Stride Reprogramming to Enhance the Performance of SSDs

Congming Gao, Jiancong Zheng, Xufeng Yang, Qiao Li, Jian Chen, Tianyu Ren, Zheng Wan, Yina Lv, Xin Xin, Min Ye, Chun Jason Xue, Jiwu Shu

2026年份

摘要

NAND flash-based SSDs enforce strict sequential page programming within blocks, primarily due to reliability constraints. However, enabling reprogramming of pages after initial writes could unlock significant optimization opportunities, such as accelerating valid page movement during garbage collection or program operations. Toward this objective, prior work has explored reprogramming. Nevertheless, its effort faces a critical limitation: reprogramming is restricted to a small subset of wordlines per block (i.e., stride), rather than extending to all WLs per block. In this work, we advance beyond this constraint with LOONG, a method that expands the reprogramming stride to encompass entire blocks. LOONG introduces a long-stride reprogram operation, which spatially decouples programming steps. First, all WLs in a block undergo sequential programming in SLC mode to maximize performance. These pages are then uniformly reprogrammed to TLC mode, preserving storage capacity. LOONG requires no hardware modifications and can be deployed solely via firmware updates. We evaluate LOONG's efficiency with two case studies: GC acceleration and program operation optimization. Experimental results demonstrate average latency is reduced by 37.5% and 18.1%, respectively, in these two scenarios.

问问这篇 Paper

问问你的智能体。

Lune 读过与它相关的顶会 Paper,每个回答都会注明依据哪几篇。

可以从这些问题问起

智能体调用

Lunesearch_papers

在 Lune 里问

免费开始,无需绑卡

相关 Paper

黄昏的海面,两侧是细线勾勒的悬崖