BBR breaks Web UI on Flint 3 (4.8.4)

Router
1

Hi everyone,

I’m running a Flint 3 on firmware 4.8.4.

I can successfully enable BBR via SSH (switching from cubic and setting fq as the default qdisc), and the router works fine from a networking perspective.

However, once BBR is enabled:

  • The GL.iNet Web UI becomes inaccessible / unresponsive

  • Routing, internet, and SSH still work normally

As soon as I revert back to:

  • net.ipv4.tcp_congestion_control = cubic

…the Web UI immediately comes back without any other changes.

This is fully reproducible on my setup.

Has anyone else seen this on Flint 3 or 4.8.4?

Is the UI tied to assumptions around cubic, or is there a known incompatibility with BBR + fq on this firmware?

Thank you!

2

Hi

Sorry for the late reply.
Let us have the R&D team take a look.

1 Like
4

We’ve confirmed that this is a compatibility issue and it may not be resolved in the short term.

For now, please continue using the cubic algorithm.

5

@newzoon Curious why you would want to set the router itself to use BBR?

6

Hi!

BBR uses bandwidth/RTT modeling instead of loss-based probing like CUBIC, which can improve utilization and reduce bufferbloat on high-BDP links.

7

Yes I get that, but why do you need that on the router? Changing to BBR only impacts connections outbound from the device itself and data it transmits. So changing it on the router would impact the router itself uploading things, or the data it sends you accessing the GUI? Doesn’t seem like this is useful? It does not do anything for data flowing through the router.

If you want these benefits for your machines you need to enable it on the devices behind the router, not on the router itself. Even then it only impacts the upload from that device. You would need to change the server out on the internet to improve downloads.

8

You’re right.. it doesn’t change congestion control for transit traffic. I enabled BBR mainly for testing when the router itself is the TCP endpoint (e.g. running speed tests from the router, firmware/opkg downloads, or any TCP service on it) and to compare how BBR vs CUBIC affect the WAN egress queue under upload saturation. I’m not expecting it to improve performance for clients behind the router, just to observe behavior at the bottleneck.

1 Like