Hi everyone,
!!! REBRAND TO DARKMOON !!!
I’ve been working on a project to slim down the footprint of the excellent cake-autorate tool, and I think it’s ready for some outside eyes.
First, a huge shout-out to @Lynx
The original works great, but I wanted to see if I could make it even leaner for routers that are a bit tight on CPU or RAM. I’ve spent the last few weeks rewriting the core logic from scratch in C .
What’s different in this version:
No more fping dependency: I wrote an integrated pinger using raw sockets. It’s more precise and uses fewer resources.Netlink instead of tc commands: The daemon now talks directly to the kernel to change rates. No more constant process-spawning (fork/exec) every time the bandwidth shifts.Lean & Mean: It’s designed to be a "set and forget" background service with a very flat CPU profile.
I need your help! Since I only have one device to test this on, I’m looking for some brave souls to give it a spin on different hardware and connection types. If you’ve been using the original script and want to see if the C version feels snappier or helps your CPU load, I’d love your feedback.
Changes at a glance:
Dropped fping and bash dependencies.
Rewritten pinger for higher precision.
Optimized sysfs and Netlink interactions to save CPU cycles.
Current version: v1.0.0-Darkmoonhttps://github.com/kamikaonashi/openwrt-package-cake-autorate-reborn/releases/tag/v1.0.0-Darkmoon
If you decide to try it out, please let me know your router model and how it handles your connection!
Cheers!
Some screenshots of the UI
13 Likes
updated the link for more architectures and compiled for ipk and apk format
updated link for v1.0.1
3 Likes
Lynx
February 27, 2026, 11:03am
3
I've always thought it would be nice to have an implementation of cake-autorate in C.
I have a couple of questions.
Firstly, how much of the original cake-autorate logic have you reproduced in terms of the overall logic concerning shape rate adjustments and reflector rotation, etc.?
Secondly, am I right in thinking that at the moment you have implemented regular ICMP pings rather than ICMP type 13 pings for true OWDs in each direction? I switched over to the latter some time ago for my own connection. I first pull in one of the big reflector lists from: https://github.com/tievolu/timestamp-reflectors then randomize, then work on those using the reflector rotation logic.
Most of the important cake-autorate logic is already ported. The parts that decide when to turn the shaper up/down and by how much are implemented (delay/bufferbloat detection, load detection, the smoothing/EWMA stuff, and the up/down adjustment rules). Reflector handling is also there: I keep multiple reflectors active and I rotate/replace them when they start behaving badly, similar to the script’s health/rotation idea.
Yeah, right now it’s just normal ICMP echo ping, and I approximate one-way delay as “RTT / 2”. So it’s not true per-direction OWD yet. But I will take your feedback and will implement ICMP type 13 soon
Really appreciate your work and feedback. I'm kinda a noob and new to Openwrt, since i mostly developed android roms before
1 Like
Lynx
February 27, 2026, 3:50pm
5
Looking forward to trying this out on my 4G connection soon, if not this weekend.
jpman
February 27, 2026, 6:06pm
6
Cake-autorate (along with adblock-lean) has provided a massive improvement for my 4g connection. A more efficient C version is exciting news! I will try the v1 release over the weekend on a spare Archer c59v2 running 24.10.5. Should be a close-to-worst-case performance test for cpu/ram.
1 Like
appreciate any feedback if you are able to test it
I just pushed an update v1.0.1 where i removed floating point math for lower end devices. hope it works outs fine
2 Likes
I'm quite inexperienced when it comes to OpenWRT after taking the jump from Stock 'Forked' OpenWRT but I've added this for my 5G NSA connection.
Error below I've rebooted the router also; any suggestions?
"RPC error: RPC call to luci/setInitAction failed with error -32002: Access denied at handleCallReply (http://192.168.8.1/luci-static/resources/rpc.js?v=26.043.84860~6cf8bf9-1772222829:11:3 )"
JackH
February 27, 2026, 9:33pm
10
This is so cool! Thank you, and welcome to the forum kamikaonashi!
1 Like
Try this via ssh
cat > /usr/share/rpcd/acl.d/luci-app-cake-autorate-reborn.json << 'EOF'
{
"luci-app-cake-autorate-reborn": {
"description": "Grant access for luci-app-cake-autorate-reborn",
"read": {
"uci": [ "cake_autorate_reborn" ],
"ubus": {
"service": [ "list" ],
"luci": [ "getInitList", "getInitStatus", "setInitAction" ]
}
},
"write": {
"uci": [ "cake_autorate_reborn" ],
"ubus": {
"luci": [ "setInitAction" ]
}
}
}
}
EOF
/etc/init.d/rpcd restart
2 Likes
Seems to have done the trick, great work!
1 Like
How is it, any improvement over the scripts on cpu side? That would interest me
Lynx
February 28, 2026, 1:21pm
14
So I've got it running on my RT3200 with the LuCI fix you reported above.
But the bandwidth adjustments are very slow.
Any logging options so we can see what's going on?
I don't have any comparisons evidential anyway; but from what I've noticed I've gained about 20mib due to switching to this.
I'm happily willing to share information if you have the commands; but me being a novice the above is what I've immediately noticed.
Lynx
February 28, 2026, 1:25pm
16
Are you seeing regular bandwidth adjustment?
As above I'm quite new to this; but Luci App doesn't seem to show anything happening afaik anyway but this below shows something is happening.
Summary
root@OpenWrt:~# tc -s qdisc show dev ifb-wwan0
Do throw some ssh commands my way and I'll do what I can.
Lynx
February 28, 2026, 1:34pm
18
while true; do clear; tc qdisc ls | grep cake; sleep 1; done
Now run speed test - do you see rates change much? Ctrl-C to stop.
My rates change about every 30 seconds or so.
2 Likes
Started the command, loaded up Speed test then ran it below I'm currently on WiFi and another device is currently watching YouTube
SpeedTest
root@OpenWrt:~# while true; do tc qdisc ls | grep cake; sleep 1; done
jpman
February 28, 2026, 5:50pm
20
My initial experience on a TP-Link Archer C59v2 is promising. For this test, I’m leveraging a 5ghz WWAN connection from this router to an LTE hotspot device. My config is here (only what I changed from defaults):
config cake_autorate_reborn 'primary'
option enabled '1'option dl_if 'ifb4phy0-sta0'option ul_if 'phy0-sta0'option min_dl_shaper_rate_kbps '8000'option base_dl_shaper_rate_kbps '24000'option max_dl_shaper_rate_kbps '60000'option min_ul_shaper_rate_kbps '1500'option base_ul_shaper_rate_kbps '3000'option max_ul_shaper_rate_kbps '8000'option connection_active_thr_kbps '1000'
Under a speed test load, I’m initially seeing about 80% cpu usage on htop at around 32mbps, which decreases into the 50% range as cake autorate reborn throttles my connection. With the default 20hz 6 pingers and .3 ping interval, this represents a significantly lower cpu usage over cake_autorate, which is in the 200-300% cpu usage range in similar conditions/settings on this single core router. I typically run 10hz or 5hz settings.
I am seeing rate changes about once per second when running
while true; do clear; tc qdisc ls | grep cake; sleep 1; done
Outside of periodic speed tests, I’m not sure how else to verify that cake_autorate_reborn is managing the connection as effectively as expected. I will run it for an extended period of time and see how the network feels when using it. Happy to perform any other tests as requested. Enjoying your work so far!
1 Like
