Cake-autorate implentation status

[hatcreek68]

Hello,
I'm in the planning phase to use openmptcprouter to bond a 5G Calyx hotspot and Starlink. I have used both of these independently and they both benefit from cake-autorate.

I see a recent post on the cake-autorate (https://forum.openwrt.org/t/cake-w-adaptive-bandwidth/191049/637) that suggests cake-autorate is integrated into openmtcprouter.

Main question. Does this mean openmtcprouter users do not need to install cake-autorate separately? Does it control each WAN link separately? If this is in the documentation already feel free to send me there.

[AKHwyJunkie]

I'm not the dev, but I've played around with SQM. It does do adaptive bandwidth management for each independent WAN link separately. I don't believe you'd need to install anything else to take advantage of it.

My personal experience with it was mixed and I've since disabled it. On variable connections (i.e. non CBR links), the "adaptiveness" can sometimes lock speeds well below average or current rates. This can be a net negative if your links feature high variability. (e.g. My Starlink and LTE connections can vary by more than 100mbit at times.) Generally speaking, I've found the impacts of uplink saturation to be less with aggregated links compared than with a single connection.

[hatcreek68]

That sounds like good information, thank you.

The rest of my hardware will arrive soon for testing. I don't have very good line of sight to the T-Mobile home internet tower so I get a thousand millisecond or more latency without throttling. Cake auto rate worked really well to control the latency to under 60 milliseconds without sacrificing too much bandwidth, so I was really happy with it.

At any rate, once I'm able to test, I'll see if my results mirror yours or not.

[AKHwyJunkie]

You'll be happy to know you can do the SQM autorate on a per connection basis, too. So, if you find it helps your LTE connection, you can keep it on there and then let your Starlink run full throttle for maximum throughput benefit. In my dealings with Cake and FQCodel, I've found it more important to achieve a good user experience than A-grade bufferbloat tests.