[RT-AC86U] I built cake!!!

[robcore]

It's been a long week, but I finally got it working!

...Okay I haven't hit "apply" yet for fear of a nuclear explosion (and next to 100% asp illiteracy), but the queuing discipline itself works when manually applied. I poured over Eric's history to ensure that it has hooks in all the right places to act as a legit option alongside sfq and the codels.

Anyone want to work on this with me? Downsides: I'm a kernel/project hacker at best, am super messy, and suck at git. Upsides: I am completely willing to break my own hardware but no one else's, am just familiar enough with universal programming techniques to hack together whatever I need, and will not stop until I accomplish my goals.

Check out the screen shot!

 

[Vexira]

How much of the features of cake work, as in how complete Is it, in exited to hear this news.

 

[skeal]

Wow! This sounds cool!

 

[RMerlin]

BTW, don't get too excited guys. This requires the 4.x kernel, which means only possible with the RT-AC86U.

 

[skeal]

BTW, don't get too excited guys. This requires the 4.x kernel, which means only possible with the RT-AC86U.

Ah! I see, nice little catch for non-4.x builds.

 

[Stephen Harrington]

Potentially pretty exciting for us 86U (and later when they come) users though!
Well done @robcore !

Does make you wonder why Asus haven’t gone down this path themselves and continue to spend their resources instead on other marketing fripperies (grumble grumble).

StephenH

 

[Find the Door]

Really intrigued with this here. Willing to help anyway I can - not technically able to but am by way of being able to test.

 

[Vexira]

BTW, don't get too excited guys. This requires the 4.x kernel, which means only possible with the RT-AC86U.

That's not an issue for me since in going to replace my 2.x kernel unit some time in the next few months.
Im waiting to see if 86u availableity improves, or the ax 88 is supported at some point.

 

[Vexira]

Just a curiosity, is it the full featureset of cake or is it just patch if you don't mine me asking, as in are all of the cake features enabled or only some?

 

[Vexira]

curios how would we get this to work if one was to for example to purchase an 86u or ac-88u if its supported?

 

[robcore]

Just a curiosity, is it the full featureset of cake or is it just patch if you don't mine me asking, as in are all of the cake features enabled or only some?

Sorry I've been busy lately. It's the full feature set. I built it into the kernel using the module that Eric Taht(sp?) maintains, and replaced iproute4.3 with the latest upstream one (took some tinkering).

So far, my biggest hurdle is finding out a) enough information about the relatively new per-host/per-mark features it has to b) actually integrate it into anything configurable enough to be worth it. As it stands right now, the ifb/imq stuff seems to have a lot of broadcom magic that I've not had the chance to look into enough .

Ps. @RMerlin your fq_codel integration/faketc/realtc work is ridiculously brilliant. Your hooks are perfect and even without cake I'm getting "A" bufferbloat ratings. I fear that at this stage of its development, cake hasn't been quite worth the hassle for the hype it gets.

 

[Vexira]

Sorry I've been busy lately. It's the full feature set. I built it into the kernel using the module that Eric Taht(sp?) maintains, and replaced iproute4.3 with the latest upstream one (took some tinkering).

So far, my biggest hurdle is finding out a) enough information about the relatively new per-host/per-mark features it has to b) actually integrate it into anything configurable enough to be worth it. As it stands right now, the ifb/imq stuff seems to have a lot of broadcom magic that I've not had the chance to look into enough .

Ps. @RMerlin your fq_codel integration/faketc/realtc work is ridiculously brilliant. Your hooks are perfect and even without cake I'm getting "A" bufferbloat ratings. I fear that at this stage of its development, cake hasn't been quite worth the hassle for the hype it gets.

That's awesome news, i Think that it needs more tweaks that's why I've seen articles about the pure Linux versions being amazing, so O guess it's possible that a few understand the hood tweaks need to be made.

Due to my router being too old in kernel version I find fq codel to be not as good as it would be on a unit like the 86u

 

[Sinner]

That makes me curious. Is there the full fq-codel in the 86u on merlin since it wouldnt need to be trimmed down?

 

[RMerlin]

That makes me curious. Is there the full fq-codel in the 86u on merlin since it wouldnt need to be trimmed down?

Depends on your definition of "full". Actual complete support requires driver-level support as well, which is something not supported by Broadcom's driver. However such driver-support is only critical if running at very high speeds (i.e. 1 Gbps).

 

[Sinner]

Depends on your definition of "full". Actual complete support requires driver-level support as well, which is something not supported by Broadcom's driver. However such driver-support is only critical if running at very high speeds (i.e. 1 Gbps).

Ahh ok. Works for me. I must say ive been stellarly happy with 384.5 with my custom freshqos.

The concept of higher cpu performance on 86u and the ax series for qos and vpn is of interest but no need at this time.

 

[Vexira]

Depends on your definition of "full". Actual complete support requires driver-level support as well, which is something not supported by Broadcom's driver. However such driver-support is only critical if running at very high speeds (i.e. 1 Gbps).

Is that the same case for cake as in no driver level support?

 

[Vexira]

Sorry I've been busy lately. It's the full feature set. I built it into the kernel using the module that Eric Taht(sp?) maintains, and replaced iproute4.3 with the latest upstream one (took some tinkering).

So far, my biggest hurdle is finding out a) enough information about the relatively new per-host/per-mark features it has to b) actually integrate it into anything configurable enough to be worth it. As it stands right now, the ifb/imq stuff seems to have a lot of broadcom magic that I've not had the chance to look into enough .

Ps. @RMerlin your fq_codel integration/faketc/realtc work is ridiculously brilliant. Your hooks are perfect and even without cake I'm getting "A" bufferbloat ratings. I fear that at this stage of its development, cake hasn't been quite worth the hassle for the hype it gets.

Maybe @FreshJR might have an idea about the marks sounds like traffic classification to me or a form there of.

 

[Sinner]

Sorry I've been busy lately. It's the full feature set. I built it into the kernel using the module that Eric Taht(sp?) maintains, and replaced iproute4.3 with the latest upstream one (took some tinkering).

So far, my biggest hurdle is finding out a) enough information about the relatively new per-host/per-mark features it has to b) actually integrate it into anything configurable enough to be worth it. As it stands right now, the ifb/imq stuff seems to have a lot of broadcom magic that I've not had the chance to look into enough .

Ps. @RMerlin your fq_codel integration/faketc/realtc work is ridiculously brilliant. Your hooks are perfect and even without cake I'm getting "A" bufferbloat ratings. I fear that at this stage of its development, cake hasn't been quite worth the hassle for the hype it gets.

Yeh as vex suggested @FreshJR would be the expert i think on the marks n stuff for qos etc. He actualy has built some awesome tool right into his script to tell you what the marks are for know traffic types. Its impressive

 

[RMerlin]

Is that the same case for cake as in no driver level support?

I don't know.

Sent from my P027 using Tapatalk

 

[Vexira]

I don't know.

Sent from my P027 using Tapatalk

That's understandable, still thank you.