Scalable, Ultra Low Latency and Adaptive WebRTC Streaming

https://antmedia.io/

by kerrarbone • 6 years ago

43 24 6 years ago

antmedia.io

jarym• 6 years ago

Only superficially familiar with WebRTC - what does this thing do to get the claimed 'low latency'?

jtsiskin• 6 years ago

I don’t consider 500ms “low latency”. You should be below 250ms before you start touting it as a key selling point.

dindresto3• 6 years ago

Live streaming video usually has a latency of 6 to 30 seconds, so in that context anything below a second is very low latency

Edit: the topic of my bachelor's thesis included low latency live streaming, so I hope I have at least some knowledge :)

sippingjippers• 6 years ago

You are extremely lucky if you get a contemporary online livestream under 10 seconds, this includes AFAIK all of YouTube, and the vast majority of HLS-based streams - which are often 30 seconds latency or worse.

Getting buffers down to 500ms is already pretty insane assuming reliability doesn't suffer. 250ms is well within the envelope of a single drop on a 3G network killing playback

hmtdmr• 6 years ago

According to many sources, sub-second is considered to be ultra low latency. 500ms is good for now but 250ms would be great.

CharlesW3• 6 years ago

I'm interested in this too, and why the "low latency" mode is enabled in everything but the community version.

eps• 6 years ago

The same reason they ask for the monthly fees for a self-hosted solution. They really want to be paid.

dep_b• 6 years ago

WebRTC is low latency by itself, otherwise you couldn’t have a video call. The challenge is to serve one WebRTC stream to 1,000,000 viewers while maintaining low latency

noen• 6 years ago

Quite simply, this isn't low latency.

The way you drop latency is by reducing or eliminating buffers, that's pretty much it.

There's a trade-off with h264 and hevc/265, as with most video codecs - the larger the buffer you have, the more delta frames you can have between key frames.

This means lower bandwidth and better compression.

So it's a constant struggle between compression and latency.

The lowest latency that webrtc can get is about 4ms one way, 8ms round trip, at least based on the project I started to do realtime steaming with https://github.com/3DStreamingToolkit/3DStreamingToolkit

We proved you could do round trip times in the real world from cloud to customer in under 25ms.

But it comes with a lot of tradeoffs and cost and complexity to make it work.

Google Stadia fundamentally uses webrtc under the covers, although they use QUIC instead of TCP/ICE.

If you ripped out all of the congestion management from the video engine in webrtc, you'd reduce another several ms roundtrip at the cost of basically no network resilience.

victor106• 6 years ago

I always wondered what optimization techniques Zoom applies to provide a smooth experience. It’s one of the best live streaming products I have used

CharlesW3• 6 years ago

Related: Low-Latency HLS (an extension of HLS, invented by Apple) is now part of the IETF HLS specification.

[1] https://www.wowza.com/blog/ietf-incorporates-low-latency-hls... [2] https://www.wowza.com/blog/apple-low-latency-hls

dindresto3• 6 years ago

I was very surprised to hear that Apple actually acted upon some of the community feedback. In particular, LHLS was supposed to rely on HTTP/2 Push (leading to issues with most CDNs) while the community solution makes use of HTTP/1.1 Chunked Transfer.

- https://github.com/video-dev/hlsjs-rfcs/blob/lhls-spec/propo...

- https://developer.apple.com/videos/play/wwdc2020/10228/

Tepix2• 6 years ago

Does the nginx RTMP module support LHLS already?

ddorian43• 6 years ago

Isn't nginx-rtmp unmaintained for several years ?

mekya2• 6 years ago

CMAF support is coming in Ant Media Server v2.2. It supports both HLS and Dash in a low latency way.

dindresto3• 6 years ago

This appears to be a fork of the Red5 media sever, looking at the Github repository. Would be interesting to know why this has been forked instead of contributing. At first glance, Red5 still seems to be actively maintained.

- https://github.com/ant-media/ant-media-server

- https://github.com/Red5/red5-server

mekya2• 6 years ago

This is another product than the Red5. There are too many additions to the Red5.

In the community edition there are MP4 Recording, IP Camera Support, HLS Support, WebRTC Ingestion, web panel, etc.

In enterprise features, it's totally different than the Red5. There is no single line of source code of Red5 in the enterprise features of the project. Everything is implemented from scratch.

erling• 6 years ago

Half a SECOND does not sound like “low” latency, much less “ultra low”...

Iwontgo• 6 years ago

As I know sub-second is considered as a Ultra Low Latency for now... 250ms would be great

rhengles• 6 years ago

I built a project for my company a little over a year ago, the project is live now. It's a recruitation tool, where the candidates fill out a questionnary with several questions and three types of question: multiple choice, text typing and video recording.

When searching for a tool to use to record the videos, I was very happy to find this open source server, so I wouldn't be locked to some service provider. The documentation is quite good and I didn't have difficulty integrating it to our project, despite the fact that I'm mostly a front-end and NodeJS developer.

Now that the project is live, we are having a few problems with some users. The videos that our users record are stored and watched later by specialists, we don't watch them live. But after the user start and stop recording, sometimes the .mp4 video file is not found on the server. But it is very nice that Ant Media has a admin page with CPU and memory usage, and it also reports when there's been a server crash.

It asks my email address and the support service receives the log files and reply at most a few days later. We're in contact right now, I'm sending them all the information I can to help resolve this issue. Of course we tested the integration ourselves, with 5 concurrent users recording, but not a single one of our tests has failed the way our users are reporting.

Scalable, Ultra Low Latency and Adaptive WebRTC Streaming

Related Stories