[aparsons]

Because Google turned the computers into a commodity. These employees swap out broken machines for the most part, but the software stack is so redundant that you can pull out a machine running a key job and the scheduler will simply retry it somewhere else. Similarly for files with GFS. All machines have their own battery so the possibility of the whole thing going down is slim. It’s beautiful how they’ve evolved from running on Sun mainframes in the early part of the century. The workers at the datacenters basically do mindless labour, so aren’t paid like a typical sysadmin.