-
Notifications
You must be signed in to change notification settings - Fork 82
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
RT-AX58U Cake and FlexQoS not running due to Hardware acceleration runner being enabled #615
Comments
Discussion on SNB fourms: https://www.snbforums.com/threads/flexqos-settings-problem.90668/ |
Hi, |
Opened the browser console instead of web inspector: SearchSuggestionController found an unexpected string value: HTTP request timeout SearchSuggestionController.sys.mjs:645:17 |
console in multi process SearchSuggestionController found an unexpected string value: HTTP request timeout SearchSuggestionController.sys.mjs:645:17 / |/ / / () ___ |
https://github.com/Last-times/CAKE-QoS-Script-OpenWrt/blob/main/cake.sh Maybe it's enough to remove the Asus cake qos and use another script. Adapt it so that you can set values in the GUI. Asus cake and adaptive qos is terrible. It doesn't work properly or doesn't work at all. The only improvement I see when using qos is that the dots are close together from the ping to bufferbloat. https://openwrt.org/docs/guide-user/network/traffic-shaping/sqm By setting cake->piece_of_cake on a 500/200 connection identical to mine. My friend achieved +1ms download and +0ms upload. Setting the download to 490 and upload to 190. It would be enough to give 480mbps and he would probably get +0ms. |
Did you ever install CakeQoS webgui it lets you adjust setting it’s not standard to amtm. https://www.snbforums.com/threads/cakeqos-merlin-v2-1-1.74309/ |
In asus cake qos merlin sucks just like flexqos, adaptive qos. This qos is broken. Qos must work identically to SQM in OpenWRT. There qos works perfectly. |
@RT-AX58U-B260:/tmp/home/root# tc qdisc This is with qos adaptive I'll post another with cake on. |
@RT-AX58U-B260:/tmp/home/root# tc qdisc This is with qos cake enabled. |
@RT-AX58U-B260:/tmp/home/root# tc -s -d class show dev br0 Maybe this command is not compatible. hm. |
@RT-AX58U-B260:/tmp/home/root# tc qdisc This is with qos off. |
DJones on the snbforum asked me to post here also. DSL-AX82U (Not v2). Bufferbloat tests performed via https://www.waveform.com/tools/bufferbloat This firmware changed the status of Flow Cache when Adaptive QoS is turned on, keeping it enabled, which breaks FlexQoS: When CakeQoS-Merlin v2.1.2 is running the Tools page reports Archer Disabled, Flow Cache Disabled, When FlexQoS v1.4.3 is running the Tools page reports Archer Enabled, Flow Cache Enabled All previous versions of the firmware disabled Flow Cache as soon as Adaptive QoS was enabled. The solution when FlexQoS is enabled is to disabling flow cache via SSH-
Bufferbloat test results then show FlexQoS working as expected : |
@Yakumounr |
That’s very broken. How about:
|
Enabled QoS cake; router reboot. Disabled runner from ssh; router reboot. ASUSWRT-Merlin RT-AX58U_V2 3004.388.7_1-gnuton1 Mon Jun 10 09:50:03 UTC 2024 ASUSWRT-Merlin RT-AX58U_V2 3004.388.7_1-gnuton1 Mon Jun 10 09:50:03 UTC 2024 ULIF='eth0' @RT-AX58U-B260:/tmp/home/root# tc qdisc |
Anyways I don't want to mess with their internet too much during the day so I'll follow up with anything you'd like to try later. Edit: noticed I didn't type qos stop the first time. Tried again to be accurate. and redacted the username. @RT-AX58U-B260:/tmp/home/root# /tmp/qos stop ULIF='eth0' |
Next time you want to test:
I would stop using the Cake add-on for testing. For Gnuton’s purposes, just need to prove if the default Cake implementation works or not. |
Uninstalled CakeQoS webgui as recommended.
|
If there are no errors from |
|
Ok, I missed the /dev/null silencers reading on my iPad. So gnuton must not build CAKE on this router. |
That would explain a lot. Thanks dave14305 I appreciate your help identifying the problem. |
under release/src/router/rc/qos.c I’m seeing the same null pointers I was getting so maybe it’s not a case of cake not being installed. ` #ifdef HND_ROUTER int start_cake(void)
} @RT-AX58U-B260:/tmp/home/root# cat /tmp/qos case "$1" in Upload
Download
stop) |
Something seems wrong with tc qdisc and I’m going to chalk it up to my initial concerns that hw acceleration runner would not turn off in 388.7.. will test 388.8 on gnuton tomorrow. For reference I opened another discussion here trying to understand how cake was built. https://www.snbforums.com/threads/compiling-of-cake-for-gnuton-build.90987/ [CODE]@RT-AX58U-B260:/tmp/home/root# tc qdisc @RT-AX58U-B260:/tmp/home/root# tc qdisc add dev eth0 root cake diffserv3 bandwidth 2355kbit overhead 18 mpu 64 atm nat dual-srchost |
Here’s the diff of both config_base.6a.67XX files comparing V1 to V2 of rt-ax58u what jumps out particularly is codel is not set on v2. “#CONFIG_NET_SCH_CODEL is not set” |
config.gz file from the router 388.7: "# CONFIG_NET_SCH_CAKE is not set" which is true at the time of checking. I'm not sure if that changes automatically at build of the firmware or on the fly. Edit: it's not set. That answers that I guess. @gnuton |
Also not set in beta 388.8 "# CONFIG_NET_SCH_CAKE is not set" |
Fixed 875b94d |
Wonderful once the build is released and I've got a chance to test it I'll close or inform you of any other issues. Thanks appreciate it. |
@Nullvoid3771 if you like you can test a snapshot with changes here https://github.com/gnuton/asuswrt-merlin.ng/actions/runs/9945740113 |
snapshot of release-rt-ax58u_v2-default (388.8 beta1) Enabled cake and noticed right off the bat Archer is still stuck on so I think that's the crux of the problem. Need to figure out why archer can't be forced off or automatically switched off when cake is enabled. I do notice with cake on tc qdisc shows PFIFO_FAST instead of sfq, htb, or fq_codel maybe a result of cake not working. In other models “#CONFIG_NET_SCH_CODEL is not set” is normally set. |
config of 388.8 beta1 /rom/etc/buildprofile (96756GW) the diff between the rt-ax58u v1 & v2 are the same. Here's a diff between build profiles for the gt-ax11000 merlin 388.8 vs rt-ax58uv2 gnuton 388.8 likely not relevent but I wanted to see the difference between devices. |
Here is the relevant code for qos cake I believe. github does a horrible job pasting code. release/src/router/rc/qos.c
}
release/src/router/www/QoS_EZQoS.asp
var qos_xobw_orig = parseInt(httpApi.nvramGet(["qos_xobw"], true).qos_xobw); //shapping UL speed /PrimaryWAN var bwdpi_app_rulelist = "<% nvram_get("bwdpi_app_rulelist"); %>".replace(/</g, "<");
|
@gnuton might need to add asuswrt-merlin.ng/release/src/router/rc/sysdeps/init-broadcom.c Lines 12156 to 12166 in 875b94d
|
Least I’ve got plenty of good eyes on this. Anyways I’m going to tap out unless you need me to test. I’m not qualified enough to explore this code lol. @gnuton |
@Nullvoid3771 can you please specify the version of your router? Test images https://github.com/gnuton/asuswrt-merlin.ng/actions/runs/10106108189 |
I own a rt-ax58u v2 I use remotely at my fathers house, and I own a gt-ax11000 and a rt-ax58u v1. So I can test between rt-ax58u v1&v2 In the context of this topic I’m only working on the rt-ax58u v2 because the v1 works fine. I’m away until likely Monday so I can’t check the test image unfortunately until I’m home. |
Hey I know this issue is about AX58U V2 but with the AX82U V2 I also can't get CAKE to work (see #496, still occurs with 3004.388.7_1-gnuton1).
so I guess CAKE is not actually built into the firmware? |
@RT-AX58U-B260:/tmp/home/root# cat /tmp/qos case "$1" in Upload
Download
stop)
Summery: -No errors in syslogs Cake is enabled and running without error, and Hardware Acceleration is off. I'm not sure if it's working going to have to do some bufferbloat/speedtests in person, doesn't work too well over vpn will likely on friday. But everything looks good from the logs. @gnuton Thanks for getting this fixed appreciate it. |
Seems to be working although I'm not super impressed with how well cake is handling it. Their isp plan is around 18 down 3 up on cable internet (just a slow plan not rural) anyways it doesn't seem to improve the bufferbloat even lowering it down to 5 down 2 up it still shows the speedtest hitting 17 down so I'm not sure if it's working. Maybe someone else running a test can confirm it. |
This is the kind of result I get when with cake on an AX82U v2. Obviously it is not working. |
I haven’t done any testing on the ax82u so is possible it might be having the same issues as the rt-ax58u v2. The current snapshot of 388.8 beta 1 has significant changes to get it functioning. Although in the 388.8 beta I’m not sure if the ax82u has the changes yet. Specifically looking to see if people are experiencing the same kind of bufferbloat issues unable to get it adequately setup to a A or A+ score. If it can get working on one device then more devices should be easier to fix. |
I might bring my rt-ax58u v1 to the location to see if it handles Qos cake better and if their is a noticeable difference. |
Cake has worked on the DSL-AX82U v1 with Gnuton's stable builds since 386.03_2-gnuton1 which was the first release including fixes Gnuton had to add specifically to enable it's support, A+ bufferbloat tests only with it enabled. It has worked through all stable builds since for the v1 up to and including 3004.388.7_1-gnuton1, I have yet to try 388.8 yet as I'm waiting for at least a beta version before considering flashing. The v2 is a different matter, I'm just affirming that there is no issue with the v1. |
Upon upgrading to 3004.388.8_2-gnuton1, cake still doesn't work on AX82U v2 so I guess it's not a "5.04axhnd 6750 or 6756" device. |
Closing issue. Thank you everyone and gnuton that has helped in bringing cake support to the ax58u. I understand there might be other models that have similar issues with cake. Start a new bug report per model if that is the case as it’s likely going to involve a lot of investigation. That concludes this bug report if you have further cake related issues on the rt-ax58u make a new report. Thanks again. |
Did the latest build really fix cake for you? because it did not seem like it was working when you were doing the bufferbloat test last week. I think you should be able to control cake with the cakeqos-merlin addon (if cake is indeed running). |
It took some time playing with the settings it’s “good enough” I needed it to prioritize a port for voip and it does that now. Before cake wasn’t even running. The bufferbloat results I did get better to a B, but well I do think there might be an issue there I’ll let others draw attention to it. |
Router Model Affected
Models: RT-AX58U
Firmware Version Affected
388.7_1
Is this bug present in upstream Merlin releases too?
No
Describe the bug
Cake and FlexQoS doesn't work because Hardware acceleration runner won't turn off.
To Reproduce
Steps to reproduce the behavior:
Apparently on gnuton 388.7 archer(runner) doesn’t disable itself automatically which is affecting both flexqos and cake. You need to manually disable it for either of these to work however despite ssh confirming the value of the runner being disabled manually, even after a reboot the UI still shows archer:enable under tools->sysinfo.
Runner is the successor to Flow Accelerator. It's a second level of hardware acceleration that is incompatible with a lot of router features. ~Rmerlin
If flow cache is enabled and needs to be disabled:
nvram set fc_disable_force=1
nvram commit
fc disable
fc flush
To check status:
nvram get fc_disable
To undo:
nvram unset fc_disable_force
nvram commit
Disable Archer (Runner/Flow Accelerator)
nvram set runner_disable_force=1
nvram commit
To check status:
nvram get runner_disable
To undo:
nvram unset runner_disable_force
nvram commit
Reboot to take effect.
Expected behavior
Hardware acceleration is not compatible with Cake or FlexQoS. This may also simply be a issue with CakeQoS webui, but since the systools shows archer is enabled as well this appears to be a bug in Gnuton.
Screenshots
Attached.
The text was updated successfully, but these errors were encountered: