I worked for a dotcom company in 1999-2001, and we had servers at a Sunnyvale data center. I never visited but was told that our stuff was in one cage, while others held stuff for Yahoo, Pets.com, etc. In other words many of the dotcoms of the era were located next to each other. I was also told that initially we were charged by square footage used but later as the server density increased (with back-to-back servers in each rack), they changed the pricing scheme to charge based on how much power was being drawn.
I have been inside a supercomputer room, which… is not the same thing at all, really, but maybe the basic idea is, with the cabinets full of processors and the raised floor for power, cooling, etc. to run underneath.
If there really were a catastrophic power or hardware failure, though, you could restart your computation from a checkpoint. Servers, on the other hand, you do not want to go down, so a massive data center would have to rely on the data being redundantly stored across multiple centers.
That is a very cool new fact for me.
And now they charge by any number of variables, # and type of VM (per minute), PaaS usage, storage and active disk, network ingress/egress and probably a billion things I don’t know of.
I feel like we all know this, but organizationally can’t seem to achieve it: how to avoid the experience of learning how poorly written applications (some COTS) or bad data hygiene results in your your wonderfully “autoscaling” infrastructure becoming very expensive, very quickly.
I worked in a AT&T data center and this is how we were set up. The bigger customers had their own cages with anywhere from 5 to 50 or more racks. There were also rows of cabinets where customers had single or multiple cabinets in the same lineup. Some customers had their own techs who would come onsite to install and maintain equipment and some would request “remote hands” assistance to do the work. In our case the network administration for AT&T and other providers was done off-site so we mostly just did racking-and-stacking of servers, cabling, data tape changes, customer-directed troubleshooting and equipment shipping and receiving.
The most humorous (and pretty much spot-on) depiction of a data center is this scene from Silicon Valley.
Odd how John’s nametag shows a much happier young man…
Before it was bought by Switch I took a tour of the Steelcase R&D 7 story pyramid in the Grand Rapids area. It’s now the largest data center in the east iirc.
Steelcase moved out during the last recession.
Amazing building it still has in place the 71 ft spherical compound pendulum.