TY - GEN
T1 - Storage cluster for persistency, CXL pools for caching
AU - Manaouil, Karim
AU - Zhang, Ji
AU - Zhe, Yang
AU - Wang, Zhou Xing
AU - Bergman, Shai Aviram
AU - Barbalace, Antonio
N1 - Publisher Copyright:
© 2025 Copyright held by the owner/author(s).
PY - 2025/4/19
Y1 - 2025/4/19
N2 - The emergence of Compute Express Link (CXL) enables new opportunities for scaling memory across data center nodes, introducing a high-speed, coherent memory pool that multiple systems can access simultaneously. While CXL offers a promising alternative to traditional memory and storage hierarchies, its practical integration with the operating systems’s page cache remains an open question. In this paper, we investigate the feasibility of leveraging CXL for storage data caching, specifically by extending the operating system’s page cache to span CXL memory pools. We prototype a design using virtiofs and a CXL memory expander in a virtualized Linux environment, demonstrating its impact on shared file caching across multiple virtual machines. Our evaluation shows that while CXL enables efficient inter-node data sharing, its bandwidth limitations require careful consideration in cache placement strategies. We discuss the trade-offs between local memory, CXL-backed caches, and storage access patterns, offering insights into the potential of CXL to enhance data-intensive workloads in modern data centers.
AB - The emergence of Compute Express Link (CXL) enables new opportunities for scaling memory across data center nodes, introducing a high-speed, coherent memory pool that multiple systems can access simultaneously. While CXL offers a promising alternative to traditional memory and storage hierarchies, its practical integration with the operating systems’s page cache remains an open question. In this paper, we investigate the feasibility of leveraging CXL for storage data caching, specifically by extending the operating system’s page cache to span CXL memory pools. We prototype a design using virtiofs and a CXL memory expander in a virtualized Linux environment, demonstrating its impact on shared file caching across multiple virtual machines. Our evaluation shows that while CXL enables efficient inter-node data sharing, its bandwidth limitations require careful consideration in cache placement strategies. We discuss the trade-offs between local memory, CXL-backed caches, and storage access patterns, offering insights into the potential of CXL to enhance data-intensive workloads in modern data centers.
KW - Compute eXpress Link
KW - CXL
KW - Data cache
KW - Operating System
KW - Page cache
KW - Shared memory
KW - virtio
KW - virtiofs
UR - http://www.scopus.com/inward/record.url?scp=105008422799&partnerID=8YFLogxK
U2 - 10.1145/3723851.3723859
DO - 10.1145/3723851.3723859
M3 - Conference contribution
AN - SCOPUS:105008422799
T3 - HCDS 2025 - Proceedings of the 4th Workshop on Heterogeneous Composable and Disaggregated Systems - In conjunction with the 2025 International Conference on Architectural Support for Programming Languages and Operating Systems, ASPLOS 2025
SP - 22
EP - 29
BT - HCDS 2025 - Proceedings of the 4th Workshop on Heterogeneous Composable and Disaggregated Systems - In conjunction with the 2025 International Conference on Architectural Support for Programming Languages and Operating Systems, ASPLOS 2025
PB - Association for Computing Machinery, Inc
T2 - 2025 4th Workshop on Heterogeneous Composable and Disaggregated Systems, HCDS 2025
Y2 - 30 March 2025
ER -