OMB-CXL: A Micro-Benchmark Suite for Evaluating MPI Communication Utilizing Compute Express Link Memory Devices

This paper talks about Message Passing Interface (MPI) libraries utilize CXL for inter-node communication.

In the HH case, CXL has lower latency than Ethernet for the small message range with 9.5x speedup. As the message size increases, the trend reverses with Ethernet performing better in latency than CXL due to the CXL channel having lower bandwidth than Ethernet in the emulated system 2 compute node with memory expander for each node.