Photos from Bell Labs Datacenter in 1960s (2019) | Hacker News


Around the turn of the millennium I worked at a major visual effects house in LA as a sysadmin - with no mainframes in sight, we still basically had all of these roles in the shop.

There was an entire tape operations team whose job it was to load in tapes of assets from clients (textures, models, frames) for a project (IIRC they were staged on netapp filers for the most part) and load out tapes of final rendered frames (and new or transformed assets) for delivery to the next step of the post-production workflow. (Even if WAN links suitable for pushing many many terabytes of data were economical at the time - and they weren't - the internet wasn't really trusted for shipping around this data back then. The ludicrous powers ascribed to movie hackers in the 90s-2000s gives you a real glimpse of the paranoia in the industry at the executive level. Napster scared the SHIT out of them.)

The operators/ops supervisors were basically the render management department. Their role was to ensure jobs ran (i.e. to render frames) and produced (at least superficially) valid output. "Data control supervisor" didn't exist by that name but the job of managing compute and storage capacity over time definitely did. It might even have been in my department (systems) but I was just a puppy then and really wasn't paying enough attention.

Fun story, one fine spring the entire render management department was out for various reasons but an urgent job had come in - the South Park movie was (IIRC) 6 weeks from release and Cinesite (IIRC?) didn't have the capacity to render out all the final frames in time. The call went out internally for volunteers to learn how to run jobs, and I gave up a weekend to babysit renders. Every day new assets would be delivered and rendered frames picked up (on huge Ampex DST cassettes - hundreds of gigabytes each! A lot in 1999, anyway.)

My task was to get shots rendered at output resolution as efficiently as possible, and to preview each shot to make sure there were no obvious errors (black frames, e.g. - software wasn't perfect.) There was a bit of an art to it, since the frames could be rendered on almost any idle CPU in the facility via our in-house distributed queuing system (race, props to erco) and capabilities of the host systems were variable. If you weren't on top of things, the shot might get hung up on some frames that ended up enqueued somewhere shitty where they'd never finish, etc. The South Park frames were pathological - out of a desire for verisimilitude, the construction paper textures were HUGE, and everything was modeled in 3D (Alias) with lighting and shadows and very small Z depths between layers. A lot of the shots I drove were from the "Blame Canada" sequence (though without sound, I didn't know it until I saw it in the theater a few weeks later) - with a huge number of characters on screen, that meant gigs of texture per frame, as every character (and usually the background) was made of polygons each mapped with a particular paper texture. No wonder Cinesite ran out of time and had to go hit up other shops at the last minute.

In conclusion, batch processing: still a thing.

(Epilogue: sadly, didn't make it into the credits - those were already done, I guess? - but for giving up my weekendI did score an invite with a +1 to the wrap party. Isaac Hayes rocked. Mary Kay Bergman held my +1's hair while she barfed in the ladies room. And as far as I could tell, Matt and Trey never bothered to show up. A+, would give up weekend again.)