Hacker News Reader

I assumed the videos were created lazily, on first view. This article makes it pretty clear that they decided from the beginning that they needed to render all the videos in advance. I wonder how many videos they rendered which were never viewed, and whether rendering on first demand would have been viable.

View on HN

I was also expecting amazing optimization tricks, like:

- Designing an MPEG encoder which can take a couple of pictures and render them on the fly, so only saving the pictures,

- Carefully spreading the views over a period of time,

- Associating the soundtrack at view time, since the soundtrack is the same for everyone,

- Using EC2 for the purpose, deleting all vids afterwards,

But no, they just used the brute force method.

It would be a good job interview question: "How would you serve 700m videos?"

- Using EC2 for the purpose, deleting all vids afterwards

Facebook has their own infrastructure and very well capable of hosting their own. And of course deleting the videos isn't a smart idea

> Using EC2 for the purpose, deleting all vids afterwards

Facebook content isn't expected to disappear at some later point in time.

Reply on news.ycombinator.com