Is this normal behaviour for Cake on Asus? ICMP ping spiking

[gameinn]

Hello everyone

I know what I'm about to ask is going to make some people laugh but for me, everything should have an explanation and I cannot explain this problem. So I have configured Cake on merlin with more than adequate headroom (I'm talking ~80% of total bandwidth with the same issues happening even if I put it to 70%).

DSLReports and Waveform will give me A+ results... happy days. The problem I uncovered when testing was cmd ping spikes at the very start of a download and sometimes during it. My testing:

1: Ping a well known IP you know is stable. For simple sake we say Google DNS with -t command
2: Verify that pings are stable for 10 seconds
3: Navigate to fast.com (or https://openspeedtest.com) which will initiate a speed test
4: See that the pings in the cmd have jumped up when the test was initiated. Sometimes 1 ping but can last for 2 pings and very rarely 3 pings. To help explain, here is a screenshot:

5: As you can see on ping 12 when the test was started in Chrome, the ping jumped up by approximately 20ms.

Openspeedtest result:

Ping 11: Start of test showing a bump of 12ms. Ping 24, another spike when the download ended and upload got initiated.

Sometimes during the speedtest there will also be small ping spikes during the test. The reason I bring this up is the following video with timestamp:

If the timestamp doesn't work, go to about 10:22 in the video. As you can see they are running Cake on a 1Gbps line and initiating the speedtest has 0 effect. I would have expected the polar opposite wouldn't you? That cake would have stumbled on a 1Gbps line but been ok on a sub 50Mbps one.

Now this test isn't 100% accurate. I have had times when initiating the test that the icmp ping doesn't spike more than say 2ms but if you try maybe 5 times you should be able to reproduce the above.

 

[L&LD]

You must be doing this on a very old router (hardware limitation vs. the NanoPi R4S being tested).

What happens when you limit it to 50% and below?

See what FreshJR Adaptive QoS may offer.

[Release] FreshJR Adaptive QOS (Improvements / Custom Rules / and Inner workings)

the install routine didn't recognize my system as Merlin, so I had to answer the "stock or not" question. Is that normal? AC3200 w/ 384.11. What's the output of "uname -o"? That's the only check it does to determine Merlin. It should be "ASUSWRT-Merlin"

www.snbforums.com

 

[gameinn]

You must be doing this on a very old router (hardware limitation vs. the NanoPi R4S being tested).

It's an AX82U. It should be more than adequate for below 50 Mbps. I checked CPU usage when running fast.com and it doesn't go above 35% on any core (most times around 20%).

Also I realise there is a cake config menu I can install but I don't know how to do that.

 

[L&LD]

Yes, it should be. But the tri-core 1.5 GHz CPU leaves much to be desired in 2023.

Use Better Search at the top of the page to install and configure cake.

 

[Volt]

I tried to enable Cake on RT-AX56U (it has almost the same SoC as AX-82U, but with 4 cores) and was very disappointed. Bufferbloat tests were perfect, but I use another and, in my opinion, a better way to test QOS. The initial problem I tried to solve was that when I fully utilize the available bandwidth using my laptop (200Mbits/s download), sites on other devices do not open or open with a considerable delay.
I use a ping monitor app on my smartphone to see how ping times fluctuate while the speed test is running on the laptop. With Cake, I get almost the same result as without QOS, ping times increase to 1000-2000 ms, with occasional packet loses. However, with properly configured Adaptive QOS I see no or very little ping time fluctuations, with sites opening immediately on other devices while the speed test is running. Probably the lack of hardware acceleration makes Cake pretty useless on routers with relatively weak SoCs.

 

[gameinn]

I tried to enable Cake on RT-AX56U (it has almost the same SoC as AX-82U, but with 4 cores) and was very disappointed. Bufferbloat tests were perfect, but I use another and, in my opinion, a better way to test QOS. The initial problem I tried to solve was that when I fully utilize the available bandwidth using my laptop (200Mbits/s download), sites on other devices do not open or open with a considerable delay.
I use a ping monitor app on my smartphone to see how ping times fluctuate while the speed test is running on the laptop. With Cake, I get almost the same result as without QOS, ping times increase to 1000-2000 ms, with occasional packet loses. However, with properly configured Adaptive QOS I see no or very little ping time fluctuations, with sites opening immediately on other devices while the speed test is running. Probably the lack of hardware acceleration makes Cake pretty useless on routers with relatively weak SoCs.

Can you give some screenshots of your adaptive qos settings? I would like to try it.

I was about to try adaptive qos but then some trend micro agreement came up and I know from reading here that's a very bad thing so I think I'm stuck with cake for now.

 

[dave14305]

and I know from reading here that's a very bad thing

Not really.

 

[Volt]

The main factor that affects bufferbloat when using Adaptive QOS are speeds that you specify in the bandwidth settings (I would not recommend you to use the automatic setting, which is enabled by default). So, run the speed test from your router several times, then reduce the maximum upload and download bandwidth results by, for example, about 5% (you can play with this number later), and enter them in the corresponding fields. As for the order of QOS categories, place them in accordance with your Internet usage scenario (however, I've noticed that actually this order does not affect the buffer bloat tests that much).

BTW, I would not say that this agreement is a very bad thing (it does not differ much from agreements required by any antivirus software), it was discussed extensively on this forum earlier.
However, I understand your concerns, so everyone decides for himself whether to agree to it. In this regard, Cake is indeed a safe alternative.

 

[gameinn]

The main factor that affects bufferbloat when using Adaptive QOS are speeds that you specify in the bandwidth settings (I would not recommend you to use the automatic setting, which is enabled by default). So, run the speed test from your router several times, then reduce the maximum upload and download bandwidth results by, for example, about 5% (you can play with this number later), and enter them in the corresponding fields. As for the order of QOS categories, place them in accordance with your Internet usage scenario (however, I've noticed that actually this order does not affect the buffer bloat tests that much).

BTW, I would not say that this agreement is a very bad thing (it does not differ much from agreements required by any antivirus software), it was discussed extensively on this forum earlier.
However, I understand your concerns, so everyone decides for himself whether to agree to it. In this regard, Cake is indeed a safe alternative.

Thanks Volt. I tried adaptive and unless I set it up wrongly, I was getting much higher ping spikes. I also dropped to an A on bufferbloat test.

I think I will stay with cake for now. It's definitely performing admirably for stuff that actually matters.

One thing I did think about for your scenario. If my router CPU is going to about 35% when shaping 40 Mbps, it's likely very close to 100% CPU load when shaping 200. Just to see if this is the case you should shape your traffic down to like 120 just to see.

 

[Volt]

Regarding the ping spikes with Adaptive QOS, one of the possible reasons is that your ISP does not always provide stable speeds, so when you set the expected upload and download values, sometimes they just exceed the available bandwidth. I periodically noticed a similar problem with my ISP when I had high ping spikes in the evenings, and a speed test showed that my ISP was only giving me 150Mbit/s instead of the expected 200 at that time. To be honest, A and A+ is not a big difference if you have problems loading sites on other devices during high bandwidth usage like I had. For example, Adaptive QOS always gave me lower buffer bloat scores than Cake, but in real-life scenarios it was working much better.

As for the CPU load, I was getting about 70-80% CPU load with Cake. If I understand right, Cake can only run on a single core, and, unfortunately, single-core performance of our routers is pretty low by modern standards.