Cloud Exit Is Real: Why Cloud Economics Break Down at Scale
simplyblock.io
simplyblock.io
companies that were in the cloud in 2018, did not have to pay the high costs of fixing hardware that was vulnerable to spectre/meltdown. in some cases, migrating from a datacenter to the cloud was cheaper than upgrading datacenters full of vulnerable hardware.
now with everyone in the cloud, datacenters have lower prices - so its going to draw companies into datacenters. in a few years, though, theyll have to either upgrade their hardware, or move back to the cloud. its gonna cycle like this for the forseeable future.
That’s the usual comeback from the cloud true believers.
The essential message that AWS wants you to believe “Your company doesn’t have the skills to run computers, and you need no technical experts at all to use AWS”.
The cloud is where Moores Law goes to die - insanely expensive slow computers.
Sitting next to him was a C-level executive from a public sector organization in one of the Mountain states lamenting the costs associated with their cloud deployments and lack of seasoned IT pros, due to the lack of ability to hire and retain IT professionals at the low wages allowed by budgetary constraint. While they could more easily gain limited investment from the state legislature for initiatives and the consulting companies advising them on strategy, they are now hamstrung by rising costs associated with what I call 'The Timeshare Points Model' where (overly-simplistically) $1 == 1 credit today but $2 == 1 credit tomorrow, even though--traditionally--compute and RAM costs go DOWN over time, not up, for the same level of service.
It was absolutely wild to me that there are still folks who truly believe that they can somehow save money without knowledgeable and thoughtful staff on hand to make smart decisions to run workloads in a cost-conscious manner and instead trust these cloud computing platforms to just do everything for them with unpredictable and ever-increasing cost.
The service I built has 99.99% availability, most of that was achieved by putting it in K8S clusters in multiple regions behind a CDN. Only managed service I rely on is SQL with better guaranteed uptime than I have (also way cheaper than doing it ourselves).
Aside from SQL, I can lift-and-shift almost all of our services at a moments notice. The only problem would be slightly degraded database query time, but we do a lot of caching in the clusters.