[whotheffknows]

It's easier to compare the underlying unit prices of the computer resources needed. Even if the same k8s model is implemented off on multiple platforms, each of the platforms have their own features inherent or ad hoc that are subject to change at any time via internal releases.

Regardless, every cloud host company is offering their own flavoring of a k8s engine which biases their open source sdks to build the networks implementing it on your own towards having various edge cases broken or not implemented at all.

The time and resources it takes to then be sure you've implemented an opensource network and underlying compute resource model as an exact equivalent for calibrated comparison is actually where the large cost of this lies, and that output of that comparison is bound to be different model by model which is designed to be unique per company design.

Tldr the real cost is the amount of time it would take to do this per cloud implementation when the results aren't scalable to other architectural implementations and varying workloads they will support.

I was given the task of doing this with effectively unlimited funds across aws, Google cloud platform and azure with a platform to support one million uses after three months of getting terraform to roll out one piece of it for google which intentionally made it lacking to bias you towards using GKE, we ended up with one implementation of GKE with failover in multiple zones and increasing feature lock-in but only security as a feature lock-in after a year, so we also had to pay for security training pentests and reviews.

The cost of all of this is not likely to provide more benefit the final comparison minus these resource ramp up costs would entail.