Distributed file systems rely on uncoordinated per-node page caches that replicate hot data across every node, wasting aggregate cluster DRAM and paying heavy coherence overhead. DPC is an OS-level distributed page cache built on CXL 3.0 memory semantics that treats all cluster DRAM as a single cache budget and enforces a single-copy invariant: each file page has exactly one owner node holding the sole DRAM copy, and other nodes reach it through CXL-based remote mappings rather than replicating it. On a multi-host CXL 3.0 emulation framework, DPC delivers up to 12.4× speedup and a 5.6× geometric-mean speedup on data-sharing workloads.

Preprint (arXiv:2604.19494), 2026.