I think you are asking about the pod management overhead on individual nodes, i.e., how much cpu/memory it takes for the per-node agent/daemon to run N pods for you on that node. This is different from the "master node" which usually runs a few master components/pods.

The resource usage depends on many factors. To name a few, the container runtime (e.g, docker/rkt) and its version/configuration, the number of the pods, the type of pods, and so on. To present users a better picture of the kubernetes per-node overhead, there is an on-going effort to create a node benchmark to let you profile the performance and resource usage on your node.

For now, you can check the k8s perf dashboard and select "kubelet perf 100" from the drop down menu to see what's the resource usage of kubelet (the node agent) and docker for managing a 100 "do-nothing" pods on the node. It should be noted that this is the best case (i.e., steady state) where no pod operations (creation/deletion) are performed. You are expected to see cpu spikes when there are multiple concurrent operations (e.g., creating 30 pods).