CXL memory sharing
CXL memory sharing allows multiple hosts to access the same device-attached memory region concurrently under a specified coherence model.
CXL 3.0 extends pooling with cross-host sharing. Hardware coherence maintains cached copies across hosts. [1] This describes cache coherence, not an application transaction or crash-recovery guarantee.
The Consortium tutorial distinguishes software-managed coherence from multi-host hardware coherence. The fabric manager selects the model. Software ownership coordination is outside the specification; hardware coherence requires device tracking, an HDM-DB mapping and a back-invalidation (BI) channel. Shared does not automatically mean hardware-coherent. [2]
The Consortium also describes a vendor prototype using a CXL 2.0 switch with software-supported sharing. This is a software-assisted prototype; it does not establish standardized hardware-coherent sharing on CXL 2.0 systems. [3]
Relations
Citations
- CXL Consortium: Compute Express Link 3.0 white paper, page 3; cross-host hardware-coherent sharing.
- CXL Consortium: Flash Memory Summit tutorial (August 2023), slides 27-28; pooled/shared FAM and two coherence models.
- CXL Consortium: XConn CXL 2.0 memory pooling/sharing prototype (30 November 2023); software-supported demonstration.
Sources & caveats
- Source synthesis of Consortium explainers, not a normative specification review, interoperability test or native CXL experiment.
- Status is unknown because these sources establish mechanisms, not current market availability. Synchronization and durability boundaries are engineering distinctions, not measured results.