[echoteecat]

In the "bad old days" a company would commission capacity studies to see how many server they would need to run the parts of their service in data centers and colos. Then they would buy racks and racks of hardware that all had tags for which part of the system they were destined for. An army of network engineers would make sure that the systems were interconnected to the right routers and switches, and that the networks were appropriately segmented. An army of sysadmins would take all of that raw hardware and would painstakingly follow endless checklists and procedure binders to make sure that each piece of hardware was correctly configured to run its part of the system.

But, with the cloud, you buy hardware, stick a base OS on it, and it becomes an amorphous amount of cpu cores, RAM, and harddrives (or at least we try to make people think it works like that). And if everything now runs in a VM, or a container, you can "just" write software that says "I need 6 database servers, 8 event bus servers, 9 back end servers, and 3 load balancers, and we'll use a CDN for all the static assets." (gross oversimplification). And if you need to reconfigure it to be 5-7-12-5-9, you just stop the old VMs and start new ones. You don't need to follow a checklist to make sure that you've correctly uninstalled MySQL 5.6.9, and need to install MariaDB 10.1.35.

DevOps is a very different mindset to traditional System Administration.