Dynamic resource allocation rethinks how additional resources like
accelerators be managed and requested in Kubernetes.

The KEP discussion showed that a longer period in alpha seems likely, so now
the (of course still tentative) plan is to keep it in alpha for three releases.

SIG Autoscaling will need to be involved, now as approver and later as
reviewer, so they get added as "participating SIG".

Finally, the goal is to start working on this in Kubernetes 1.25, so it has be
marked as "implementable".

…ocation

NVIDIA cards in MIG mode are limited by the current need to pre-partition the
hardware.

This discusses the effect of attacking the resource driver in addition
or instead of kubelet.