Lune

ICDE2026顶会

SHMemora: Protective Key-Value Store on Distributed Shared Memory

Jiajun Luo, Siyu Lin, Yunpeng Xu, Shengwei Liu, Jin Xia, Dong Liu, Zheng Liu, Huanchen Zhang, Teng Ma, Shuwen Deng

2026年份

摘要

Modern in-memory key-value stores (IMKVS) are essential for data-intensive applications but increasingly limited by memory capacity. Traditional scale-out over TCP/IP or RDMA incurs high overhead from network latency and data movement. Emerging Compute Express Link (CXL) enables nearlocal-latency shared memory across hosts, offering a promising path for scaling IMKVS. However, existing CXL-based systems prioritize performance and overlook security, leaving shared memory exposed to unauthorized access. In this work, we present SHMemora, to the best of our knowledge, the first protective multi-host IMKVS built on CXL-based shared memory in coherent Distributed Shared Memory (DSM) environments. SHMemora addresses three key challenges, including strict protection enforcement, fine-grained memory protection, and scalable metadata management. These are solved by SHMemora through Intel MPK-based memory isolation, a capability-based protection, and metadata batching, respectively. We implement SHMemora on real CXL hardware and evaluate its security through formal proofs and penetration tests, and its performance through microbenchmarks and real workloads, demonstrating both strong protection guarantees and practical scalability. SHMemora achieves an average of 3.45×{3. 4 5} \times throughput compared to Redis under multiple hosts, incurring an average of 1.68% throughput overhead for the security protection.

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