Have you considered that your opinion on the best way to serve data to a billion users might actually be wrong? That maybe Facebook actually knows what they are doing?
Not really surprising, from what I’ve heard from friends about their cowboy-style internal engineering culture, and spaghetti piled on spaghetti internal code organization, where shipping new features ASAP is privileged above all other goals.
When you have endless money to throw engineers and servers at a problem, it’s possible to kinda sorta paper over a lot of horrible broken design. Not sure whether that implies anything one way or another about whether or not they “know what they are doing”.
It sort of depends on your definition of “know”. After all, their motto is “move fast and break things.”
Certainly! But I've also done work where Humvees in the middle of Iraq use satellite internet connectivity with massive latency and the more connections it has to open to get started the slower everything gets.
Perhaps this is an advantage where latency is low so you can load parts of the page faster but I'd love to know some of the more technical reasons behind this!
[1] http://quellish.tumblr.com/post/126712999812/how-on-earth-th...
[2] https://www.facebook.com/notes/facebook-engineering/under-th...
I mean, isn't that how monolithic apps tend to happen? Someone leaves, new person comes along and adds more to the cruft.
But I don't know anyone at Facebook so I have no idea how the website is handled.
So no matter how many connections are made when visiting facebook, it usually loads fairly fast because of aggressive caching and progressive connections. Most of the time the end user shouldn't be able to tell the difference.
I suspect their mobile apps are more conservative about using fewer requests (and probably a binary protocol).
This horror out of time and space was beaten, shot and stabbed, and the developer responsible for its invention was promoted to CTO and King of the universe.
Actually it is: https://www.facebook.com/notes/facebook-engineering/bigpipe-...
Nope. Did you read through how BigPipe works? It's more than a handful; they basically load a whole bunch of mini web pages inside of another. I'm actually impressed they get better performance doing that; I would have thought condensing everything down would net much better performance. It would at least improve latency but I digress.
I like this part. As a developer I've often looked at the network usage of large websites / web applications and it's always surprising to me just how...unoptimized it is as far as network connections go.
I mean Facebook loads decently enough and all I'm just surprised the first load isn't condensed into a small, handful of network calls to save on latency.