Alta Labs just published a video explaining their CAKE implementation: https://www.youtube.com/watch?v=vxDO0flKcI0 All the best, Frank Frantisek (Frank) Borsik https://www.linkedin.com/in/frantisekborsik Signal, Telegram, WhatsApp: +421919416714 iMessage, mobile: +420775230885 Skype: casioa5302ca frantisek.borsik@gmail.com On Mon, Mar 24, 2025 at 4:39 PM Daryll Swer wrote: > I was looking at your Route10 product, but it looks to be consumer-heavy, > at this point time. I run stuff like BGP in my home network (as I have > MPLS-to-the-home delivery) and PIM-SM, would be cool if you folks supported > some DC-fabric features on Route10 in the future with FRR as the daemon or > something. > > One thing I'd like to mention, Route10 should have a good NAT > implementation to support EIM-NAT for TCP/UDP to allow P2P networking to > work over a NAT box. > Additionally, support NAT Hairpinning for user's LAN subnets to allow > intra-NAT traffic to work over STUN discovery, this eliminates TURN traffic. > Small to large-scale CGNAT deployments around the globe is something I've > been consulting for a few years now. I'm happy to potentially beta test > your products, if you'd like, for best practices conformance with very > specific RFCs. > > BQL would be good to have for long-term viewpoint, perhaps your Ethernet > driver provider, could patch it up for BQL support and that would be a good > foundation for FQ_Codel on your Route10. MikroTik has FQ_Codel, but lacks > BQL, and they refused to listen to Fran and Dave Taht, so perhaps you folks > are our only hope :) > > *--* > Best Regards > Daryll Swer > Website: daryllswer.com > > > > On Mon, 24 Mar 2025 at 20:53, Jeff Hansen wrote: > >> Frank, >> >> The hardware that Route10 is based off of doesn’t support FQ Codel nor >> CAKE at all, so everything is done in software on our 5.4 Linux kernel. It >> works great, though. In some instances it’s the only way to max out a PPPoE >> connection and have optimal latency. >> >> It doesn’t look like our ethernet driver supports BQL at all, so we >> haven’t tried that yet, but as is, it absolutely eliminates high latency if >> tuned properly. >> >> -Jeff >> >> On Mar 21, 2025, at 2:27 AM, Frantisek Borsik >> wrote: >> >> Happy to see that! Thanks, guys. >> >> Adding Jeff >> , >> Alta Labs CTO - Darryl >> >> has a suggestion how to push this further: "Maybe they can add both >> FQ_CoDel and CAKE with BQL support? How's hardware-offloading of FQ_CoDel >> looking on these “prosumer” equipment these days? I haven't kept up over a >> year on this topic." >> >> All the best, >> >> Frank >> >> Frantisek (Frank) Borsik >> >> >> >> https://www.linkedin.com/in/frantisekborsik >> >> >> Signal, Telegram, WhatsApp: +421919416714 >> >> iMessage, mobile: +420775230885 >> >> Skype: casioa5302ca >> >> frantisek.borsik@gmail.com >> >> >> On Fri, Mar 21, 2025 at 3:44 AM Jonathan Morton >> wrote: >> >>> > On 19 Mar, 2025, at 12:01 am, Frantisek Borsik via Cake < >>> cake@lists.bufferbloat.net> wrote: >>> > >>> > Should be pushed through production in day or two and they will be >>> talking about it on https://streamyard.com/watch/ubYm2AffWkYi >>> >>> this Wednesday, March 19, at 1PM EST / 12PM CST / 11AM MST / 10AM PST >>> >>> I joined the stream, and was able to ask about the throughput they were >>> getting with CAKE on their hardware. This is just for the "Route 10" >>> rather than their APs, and they reported getting about 2.5Gbps throughput >>> with CAKE enabled. They do correctly note that the hardware-accelerated >>> forwarding path is disabled for the interface where CAKE is turned on. >>> >>> Supporting 2.5Gbps is pretty good I think, and should be sufficient to >>> handle all practical Internet subscriptions that are likely to require >>> bufferbloat mitigation. For comparison, on the same call they claimed >>> about 800Mbps throughput for acting as a WireGuard tunnel endpoint. >>> >>> - Jonathan Morton >> >> >>