Lune

ICDE2026顶会

REMON: Remote External Memory Over the Network

Shiquan Zhang, Michail Bachras, Yuqiu Zhang, Yunhao Mao, Hans-Arno Jacobsen

2026年份

摘要

Modern database systems face significant challenges in managing memory efficiently, especially under conditions of resource heterogeneity, constrained local memory, and growing dataset sizes. To address these issues, we introduce REMON, a lightweight, software-based disaggregated memory system that leverages remote memory resources over the standard networking stack. REMON employs page-level memory management and efficient swapping mechanisms, enabling seamless memory allocation, release, and swapping between compute and memory nodes. Across standard analytics benchmarks, REMON reduces average latency by up to 34% over disk swapping under memory pressure and incurs less than 15% performance loss relative to fully in-memory execution when data fits in RAM. Compared to the RDMA-based remote-memory baseline, REMON is on average 11% slower but remains deployable and cost-effective on commodity cloud instances; its benefits are strongest in lowlatency, high-bandwidth in-region clusters. With its user-space design over commodity TCP/IP and cloud-native compatibility, REMON offers a versatile, cost-effective memory extension for cloud databases.

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