Messing Up Your NUMA Topology with CXL - Implications for Advanced Interconnects

Explore the advanced interconnect technology Compute Express Link (CXL) and its impact on NUMA topologies in this 42-minute conference talk. Delve into how CXL leverages PCIe physical interconnect to link various device types, enabling tailored solutions for large-scale AI systems and dynamic resource pooling. Examine the implications of CXL on memory placement, considering its hotpluggable nature and potential unsuitability for certain data structures like DMA areas. Gain insights into the creation of complex NUMA topologies through CXL's ability to connect and pool memory resources. Conclude with a brief demonstration using an emulated CXL instance under QEMU to illustrate key concepts.