4 ms·

Sounds like overkill, mate. I’ll go with your comments parent. Maybe 5 peanuts. In all seriousness, I’ll agree on cdn at 20 a month plus 3 @ 15 dollar nodes run

Sounds like overkill, mate. I’ll go with your comments parent. Maybe 5 peanuts. In all seriousness, I’ll agree on cdn at 20 a month plus 3 @ 15 dollar nodes running out app and db, let’s go multimaster with a local db node on each app server, screw the c go for a and p, and yeah a single haproxy hardware thread, ha, so make that 2 single core nodes 1gb will do fine for 300k connections per second, um you mentioned staff? For what exactly? 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

nine_k 5y ago
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.
- Dylan16807 5y ago
  > 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.
cvwright 5y ago
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?
- eyelovewe 5y ago
  You do realize that their data includes half of humanities personal lives, if you count WhatsApp and tinder. It ain’t all cat pictures, I’m afraid

nine_k 5y ago

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.

Dylan16807 5y ago
> 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.

cvwright 5y ago

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?

eyelovewe 5y ago
You do realize that their data includes half of humanities personal lives, if you count WhatsApp and tinder. It ain’t all cat pictures, I’m afraid