The Performance of the AdaptiveBweThreshold on Real Word Mobile MVCAs

[Laurenz Grote]

My Question

I have the suspicion that the AdaptiveBweThreshold is not working as intended in the open source webrtc.org implementation of Google Congestion Control. Has anybody any insights on this, has anybody else conducted a real word measurement study of the interplay of Google CC with TCP (CUBIC | BBRv1)?

My own measurements

I did conduct such a study by running Signal and Jitsi Calls (which under the hood use WebRTC + Goog CC) in parallel to a TCP (CUBIC | BBRv1) via a bottleneck. In summary, I found that the WebRTC flow is subject to starvation (gets less than 20% of available bandwidth) using a vide range of bandwidths and RTTs. The effect is even more accentuated when the call is started after the TCP flow.

As one example, see the traffic shares of the call flows against TCP BBRv1 on a 2mbit bottleneck router with Tail Drop Queueing:

With [1] in mind, I'd expect the WebRTC flows to make use of around half of the available bandwidth. Interestingly, webrtc.org implemented other values for kup and kdown, then what was proposed in [1]. The paper proposed (ku,kd) = (0.01, 0.00018), the webrtc.org uses (ku,kd) = (0.0087, 0.039) trendline_estimator.cc.

[1]: Carlucci, Gaetano, et al. "Analysis and design of the google congestion control for web real-time communication (WebRTC)." Proceedings of the 7th International Conference on Multimedia Systems. 2016.
https://mmsys2016.itec.aau.at/papers/MMSYS/a13-carlucci.pdf

[Philipp Hancke]

Stefan, thoughts?

My last state on that is https://www.youtube.com/watch?v=G2_MVsuA0CY and ... that was almost seven years ago ;-)

--

---
You received this message because you are subscribed to the Google Groups "discuss-webrtc" group.
To unsubscribe from this group and stop receiving emails from it, send an email to discuss-webrt...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/discuss-webrtc/851f1f72-2bc4-4d5b-87ba-55255657ada4n%40googlegroups.com.