How will CXL pooling and Kubernetes co-exist?
I have been reading more about the CXL 2.0 spec, specifically resource pooling. This feature enables servers to pool resources such as accelerators and/or memory that can be assigned to different servers depending on the workload. I have limited understanding of Kubernetes, but the basic idea to solve is being able to scale up or out depending on usage. Anyone have ideas about how CXL will change the landscape up on the stack?
Comments
First note that CXL 1.1 is still in the future and CXL pooling is pretty speculative. It may never happen like PCIe switching and pooling never happened.
Putting that aside, pooled resources are easier to schedule than local resources, so it should be straightforward for any cluster scheduler to allocate and assign those resources. The resulting utilization can be higher because resources won't be "stranded" on one server.
Man... failure of PCI Switching (e.g. Advanced Switching Interface (ASI)) to materialize is still a huge sadness to me. It's a shock to me how node-bound computing still is. Faster ethernet is finally semi-available, but it feels like there haven't been any real leaps or changes.