This message was deleted.

There is no difference config of livekit-server itself between oss and cloud, I think the difference is caused by the network quality between your host server and cloud node, the cloud has a algorithm to allocate a node that has best connection quality and server load for the client

Thanks for a quick answer @early-gigabyte-44411. Is there any way to confirm this? The connection on hosted oss is great over Wi-Fi (tested 3 different networks), but degrades drastically after about 10 seconds on LTE never to return to higher bandwidth. It’s looks like being throttled down for some reason. I don’t understand where the throttle comes from – network, browser, device, turn server or host?

@agreeable-minister-78501 it likely has to do with how your server is configured. I would check firewall settings on the instance/hosting provider to ensure that the ports we require are opened.

@dry-elephant-14928 Thanks for feedback. We’ve opened port 7882, which was closed due to use of 50000-60000, but it didn’t help. I looked into the issue more thoroughly and noticed behaviour: 1. When connecting via LTE from iPhone the connection degrades always after 30 seconds (not about 10 as I mentioned before). 2. After that in console on iPhone appears:

secondary PC state changed disconnected

Client starts to reconnect. After that connection is restored, but after another 30 seconds it happens again. During that time on iPhone I can see video from other call participants without any interruptions. I also tried using data channel to keep the connection and it works during the “disconnection” phase. We use LiveKit 1.3.3

The disconnected means network not stable for the rtc connection, but it will continue work until state transit to

failed

, so the video is not interrupt

@early-gigabyte-44411 That might be true, it actually starts to reconnect after state transit is

failed

. Here is full log

@agreeable-minister-78501 I have post you a test server and token, you can check it with your client

Hey @early-gigabyte-44411, @dry-elephant-14928, I’ve tested this issue on 4 different devices, 3 networks (1 wi-fi, and 2 cellular networks) and 3 servers (ours, rauber’s and LiveKit Cloud). The result is still the same: both on rauber’s and our open source server connection is being lost every 30 seconds on iOS with a cellular network. Mac & Android on a cellular network aren’t affected by this problem. Connecting with LiveKit Cloud on a cellular network works just fine. I was wondering, maybe LiveKit Cloud uses a different version of livekit-server? Is there any way to check it? Our host use 1.3.3

That's interesting.. Cloud uses the same LiveKit SFU code. However, we do enable ICELite on Cloud. That's one difference

@dry-elephant-14928 Thanks for looking into this. Can ICELite be enabled on open source version? I couldn’t find much info about it. I might also add I’ve tested the connection on 3 different iPhones (14 pro, 12, X) with iOS 16 and 15 on 3 different SIM cards (including 2 mobile service providers). Is it possible the server somehow limits the bandwidth instead of a phone?

Yup, you can enable ICELite, it's here as long as you are able to connect. the quality of connection depends a lot on your hosting provider and how they are connected to various ISPs. LiveKit Cloud runs across a number of providers that we've vetted to have performant networking and interconnects to various ISPs.

Thanks, I must have missed it in the config. We’ll give it a try and let you know how it went

hmm, that shouldn't be a major difference.. do you know how the connections have been terminated? can you enable debug on the client sdk and attach logs there?

Do you mean client JS sdk? If yes, it looks just like on the screenshot I’ve attached before I can post full log (as text) if needed in the morning (GMT+1 timezone). I’m not sure I understand the question about terminating the connections. Our sys admin was monitoring literally all data packets on the host while the issue occured, but he couldn’t find anything useful

do you have timestamps for these logs? from their alone it's not visible that they are becoming disconnected.

Yes, here is a screenshot (it’s easier to read than plaintext due to code highlight). I’ve just sent you a PM with our host url and test tokens – maybe it will be helpful to try it on your device

That's great! Can you check the users that solve the problem are connected through TURN server?

How can I check that? I’ve observed chrome webrtc internal logs and they show TURN urls for ICE candidates. Also the issue occurs again if I disable TURN in livekit.yaml config

You can check the ICE candidate grid of

webrtc-internals

tab, there will have the selected candidate pair information.

Yes, I can confirm the connection uses TURN udp

That means your client’s udp traffic for iPhone seems not stable, some ISP may block STUN/DTLS traffic, but that can’t explain your other device (android, pc) works well in same network🤔