Understanding and Profiling CXL.mem Using PathFinder#

Xiao Li et al. (UW-Madison/Beihang, Intel)

Why: CXL memory pooling is becoming widely deployed and can stall co-located apps; the community lacks end-to-end profiling, which matters for DC resource disaggregation.

What: A utility that captures CXL.mem execution end-to-end between CPU and remote DIMM for profiling, bottleneck analysis, and anomaly diagnosis.

How: Models the CXL.mem data path as an execution graph from architectural performance counters and applies graph analysis to associate CXL with competing memory flows.

Where applicable: Servers with CXL.mem; a general diagnostic tool, increasingly relevant as memory disaggregation lands.