Distinguish between NUMA memory onlined as CPU memory vs. GPU memory

[klueska]

Enhancement Description

• One-line enhancement description (can be used as a release note):
  Build a comprehensive solution to ensure Kubernetes can distinguish between NUMA nodes brought online with CPU memory vs. GPU memory (or more generally any device memory). This will likely take the form of changes to cadvisor at the lowest layers.

• Kubernetes Enhancement Proposal:

  • TBD

• Discussion Link:

  • Support TopologyManager in Large NUMA System (NVIDIA GB200) kubernetes#135581 (comment)
  • https://developer.nvidia.com/blog/understanding-memory-management-on-hardware-coherent-platforms/#how_numa_impacts_kubernetes

• Primary contact (assignee):
  @klueska, @PiotrProkop, @fanzhangio

• Responsible SIGs:
  /sig node

• Enhancement target (which target equals to which milestone):

  • Alpha release target: TBD
  • Beta release target: TBD
  • Stable release target: TBD

[ffromani]

/cc

[zvonkok]

/cc