We have a kops-managed kubernetes cluster on AWS in our company. While we are adding more and more worker nodes to our cluster, the question arises whether we should use larger worker nodes (and reduce their number) or keep adding more worker nodes of the same size.

The arguments I have in mind are:

• Too less worker nodes make it hard to handle an outage of a worker node. E.g. With only two nearly fully loaded worker nodes you can't handle an outage of one worker node. With around 20 worker nodes this should be possible.
• Performing a rolling update of the cluster takes longer with more worker nodes.
• Every worker node has a little bit of overhead caused by the operating system and processes like docker and kubelet. Using larger worker nodes makes this overhead smaller on a percentage basis.
• Having small worker nodes and large applications can lead to "one app per node", which can be inefficient in terms of resource utilization.

Are there more arguments to take account of?

And is there a break-even-point or rule of thumb?