back
8 comments
I've found BBR to be fantastic for WAN and internet use cases while having no negative impact on my traditional internal use cases. I'm glad to see fair bandwidth competition with other protocols is still a focus, though I don't see many notes on whether how much it stomps on traditional flows is any better.

I haven't really looked into BBR.Swift since I don't have many high bandwidth cases where I'm having issues or much congestion in the first place but it's good to see it being tuned there too.

AFAIK BBRv2 never made it into the mainline kernel. I've seen some people manually building and performance testing it. Curious what kernel branch/version will pick up v3 first given v1 has been deprecated according to the article. I've only personally used it on file sharing servers and have stuck with CDG CAIA Delay-Gradient for all my other hobby nodes and home routers/firewalls/dns/chat servers.
Anything that can be done will be done, in the genetic realm, in a few tens of million years give or take. The genotype is not in a hurry.
If I’m reading correctly, getting this on a local server requires kernel patches?
TCP congestion control is in kernel and QUIC (rides on top of UDP) congestion control is user space/part of each app's implementation. The former will require some updating of your local kernel.
Are there any open source QUIC implementations with BBRv3? It is hard to tell if the recent commits to Google'S QUICHE repo amount to BBRv3, or if they are somewhere between v2 and v3.
Nice work. Youtube still doesn't load any faster though...
Both the TCP and QUIC variants of BBRv1 markedly improved YouTube's experience on jittery links. It might not mean much if you already have a good broadband connection but it makes a world of difference for the hundreds of millions who only have a smartphone with cellular data.