Hacker News new | past | comments | ask | show | jobs | submit login
Scaling live streaming for millions of viewers (fb.com)
39 points by kiyanwang on Nov 10, 2020 | hide | past | favorite | 16 comments

Somewhat entertaining to notice the usage of exaggeration adjectives in this article. So they've created a broadcast system; but from the adjectives one would think they created something much more significant. Every possible place they could use an exaggerated adjective to describe the project they use an extreme form. If this tech were something new or specifically unique, maybe, but not this.

Idk, maybe I'm naive and can't see through the hype, but solving this - " We built in redundancy at every point — transport, ingestion, processing, delivery, and playback — to make sure the system could withstand most types of failure." is pretty impressive. The scale of an engineering project like that seems nuts to me.

I think it's good work, but their over glossing in their descriptions. I commented because I felt the article's adjective enthusiasm ultimately distracts from the work.

Sure, but it's Facebook so you're supposed to hate it. If it's a medium blog about Rust, we could ignore over-enthusiastic adjectives and upvote.

All that work to re-create an inferior shadow of broadcast TV from the 1970s...

Is that circuit switched or packet switched?


not a whole lot of information in this article. Maybe someone who knows better than me can say whether this is interesting or impressive? Seems like marketing material to me

They don't mention latencies. Getting below a 5 second latency is the hardest problem to solve at scale.

it's a shame multicast isn't universally supported on the Internet. It'd make it possible to broadcast a live video to millions or billions of people straight from a cell phone.

Apart from it not really being in anyone's financial interest - not content providers, copyright holders or advertisers, it would be viewed as a security risk if anyone could easily do that - e.g: https://www.theverge.com/interface/2019/4/4/18294951/austral...

The idea of using multicast in this way also ignores some of the actual good features of HLS, in particular the ability to adapt content to a specific user's device and Internet connection.

There are a few countries where multicast is usable.

It worked on 80+% of UK internet connections in circa 2008.

That was my first thought.

What's needed to enable multicast? Work by ISPs?

There is no obvious benefit to the ISP to serve multicast packets (so some kind of regulation would need to be crafted to force them to do so)

In a world where denial of service attacks are cheap and ubiquitous, my first thought is that enabling multicast would be difficult to do without reducing the cost of debilitating DOS attacks.

Well, there is.

Say I'm an ISP with X million subscribers.

Given a peering with Y GBit of throughput to someone like say, Twitch or Facebook; with multicast, I now only end up with a single load of data for each livestream coming over the pipe.

However, this does require direct peerings. Multicast over the Internet is just not a thing. This could technically be built out with IPv6, but nobody's realistically going to arrange for globally-unique multicast space allocations and convince every AS in the world to start peering multicast over BGP for general public use.

There's a certain irony to the fact that their "responsive" website doesn't scale correctly.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact