←back to thread

Netflix buffering issues: Boxing fans complain about Jake Paul vs. Mike Tyson

(www.sportingnews.com)
492 points storf45 | 2 comments | 16 Nov 24 03:42 UTC | HN request time: 0s | source
Show context
yusyusyus ◴[17 Nov 24 04:06 UTC] No.42161890[source]▶
I wrote an analysis on doing this kind of unicast streaming in cable networks a decade ago. For edge networks with reasonable 100gig distribution as their standard, these would see some of the minor buffering issues.

There is a reason that cable doesn’t stream unicast and uses multicast and QAM on a wire. We’ve just about hit the point where this kind of scale unicast streaming is feasible for a live event without introducing a lot of latency. Some edge networks (especially without local cache nodes) just simply would not have enough capacity, whether in the core or peering edge, to do the trick.

replies(1): >>42163964 #
1. ta1243 ◴[17 Nov 24 12:56 UTC] No.42163964[source]▶
Saw an Arista presentation about the increase in SFP capacity, it's Moore law style stuff. Arm based kit has a shockingly efficient amount of streams-per-watt too.

I can't see traditional DVB/ATSC surviging much beyond 2040 even accounting for the long tail.

You're right that large scale parallel live streams has only become feasible in the last few years. The BBC has some insights in how the BBC had to change their approach to scale to getting 10 million in 2021, having had technical issues in the 3 million range in 2018

https://www.bbc.co.uk/webarchive/https%3A%2F%2Fwww.bbc.co.uk...

Personally I don't think the latency is solved yet -- TV is slow enough (about 10 seconds from camera to TV), but IP streaming tends to add another 20-40 seconds on top of that.

That's no good when you're watching the penalties. Not only will your neighbours be cheering before you as they watch on normal TV, but even if you're both on the same IPTV you may well 5 seconds of difference.

The total end-to-end time is important too, with 30 seconds the news push notifications, tweets, etc on your phone will come in before you see the result.

replies(1): >>42168272 #
2. yusyusyus ◴[17 Nov 24 23:26 UTC] No.42168272[source]▶
> Saw an Arista presentation about the increase in SFP capacity, it's Moore law style stuff.

SFP itself isnt much the issue. Serdes is, and then secondarily the operating power envelope for the things (especially for the kinds of optics that run hot). Many tradeoffs available.

>I can't see traditional DVB/ATSC surviging much beyond 2040 even accounting for the long tail.

Tend to agree in well-developed infra, but rural and poorly-developed are well served with more traditional broadcast. Just saying “starlink!” 5 times in a dark bathroom won’t fix that part.

> Personally I don't think the latency is solved yet -- TV is slow enough (about 10 seconds from camera to TV), but IP streaming tends to add another 20-40 seconds on top of that.

I dont think it will get better. Probably worse, but with net better service quality. HLS/DASH are designed for doing the bursty networking thing. Among good reasons for this, mobile works much better in bursts than strict linear streams, segment caching is highly effective, etc.

But I think this makes sense: its a server-side buffering thing that has to happen. So assuming transmuxing (no transcoding lol) and wire latency are 0, we’re hitting the 1-5 seconds for the segment, probably waiting for a fill of 10 seconds to produce the manifest, then buffering client-side another 10 or so. Throw in more cache boxes and it’ll tick up more. It is quite high, but aside from bookies, i dont know how much people will actually care vs complain.