This message was deleted.

It’s how sfu works. Imagine we have 1 publisher and two subscribers. Publisher send 1MBps video stream. So input will be 1Mbps and output will be 2 Mbps because livekit server will send this stream to each participant separately

@bumpy-terabyte-27243 @loud-jewelry-13301 is right. For the large meeting use case, it’s 150 video and audio streams in and 150 * 149 streams out.

@loud-jewelry-13301 @magnificent-art-43333 Take a video conference room not so big for example, say it a 10 person room. If each of them publish one video and subscribe all of the others 9 video, how many is the publisher and subscriber in total? Say it video bit rate 1Mbps, how much is the throughput of the media server in total?

That will be 10 Mbps up stream - each person sending 1 Mbps video to the server And 90 Mbps down stream • each person subscribes to video of the other 9 people in the room. • each subscription is 1 Mbps, so each person is pulling down 9 Mbps in the down stream direction. • 10 people * 9 Mbps per person = 90 Mbps aggregate in the down stream direction.

@fancy-wire-61616 Thank you, it is the standard SFU media server logic I think. But the benchmark table really confuse me...

That is the general logic. As 150 people is not practical, i. e. you cannot really watch all 150 people, the test might have some constraints (I have to check). Also, there is simulcast. With simulcast, publishers will publish up to three resolutions at different bit rates. On subscribe side, depending on application/bandwidth, what a subscriber pulls down could be different. For example, publisher could publisher three layers at (720p at 1.7Mbps, 360p at 600 kbps and 180p @ 150 kbps). The subscriber application could be viewing only a thumbnail of this publisher's video. So, sending 720p video is wasteful. 180p video is more than enough. So, what the subscriber pulls down is only 150 kbps. In this case, all publishers could be sending ~2.5 Mbps up stream. So, the up stream number looks big or to put it differently, the down stream number does not look as big because the subscribers might be pulling down lower resolution video. From the bench mark, the important thing is that the scale that can be achieved on a given server. Packets per second/bytes per second is the important metric. Depending on application, there are a lot of combinations where the rates could be significantly different.

@fancy-wire-61616 You are right, thank you very much!