Indeed, serving is the easy part, once it's set up. Going offline when the DC goes offline can be completely acceptable.
It's the picture intake pipeline that eats most of the resources. If you only want to serve the same collection, or a collection you personally increase by 1-2 cats a day, then it's fine.
Re:The 'Fuck You' Pattern
by Dylan16807 on Monday June 28, 2021 @11:57PM
(#27671595)
> It's the picture intake pipeline that eats most of the resources.
How so? Let's say you want to be somewhat competitive with imgur and handle a million images per day, and also that the peak rate is 4x your average. That comes out to 50 images per second. A thumbnail or few shouldn't really take more than a CPU-second, so you can handle that with 2-6 cheap servers. Let alone using spare cores on storage-optimized servers. And it's next to zero database load.
You think that Facebook and Google build these huge warehouse-size data centers all over the world for fun?
You think they wouldn’t rather have the cash?
Anyway, you can do anything the hard way, but what’s your point? That serving millions of cats per second requires big investment that must be recouped? I see most of this as rent seeking, and the infrastructure is economy of scale. Maybe you use tensor flow and Kafka inside Kubernetes, I dunno…
I agree that one can expect lame websites to act lame, but it’s out of sheer churlishness rather than necessity