Cake lowers the bandwith 95% anyway?

[mutlucan]

I'm using the Cake that comes with Merlin, no additional addons. It's recommended to define 95% of the actual speed but what I see is it is done automatically. I get 100/5 with QoS off. I set 100/5 on QoS settings. When I perform a speed test, I get 95/4.8 with Cake QoS on. Also, getting A+ on everything on the DSLReports test. Why is it recommended to set a 95% limit? Or do I still need to set 95% for best results?

 

[Smokey613]

I'm using the Cake that comes with Merlin, no additional addons. It's recommended to define 95% of the actual speed but what I see is it is done automatically. I get 100/5 with QoS off. I set 100/5 on QoS settings. When I perform a speed test, I get 95/4.8 with Cake QoS on. Also, getting A+ on everything on the DSLReports test. Why is it recommended to set a 95% limit? Or do I still need to set 95% for best results?

When I was running Cake, I found that it pretty much already factored in a 10% reduction so I just set my speeds to actual tested bw results.

 

[dave14305]

At 100Mbps with 1500 MTU, and overhead of 18 (docsis), you could expect no more than 96.18 Mbps over IPv4.

Goodput = 100 * ( ( 1500 - 20 - 20 ) / ( 1500 + 18 ) )

 

[Smokey613]

At 100Mbps with 1500 MTU, and overhead of 18 (docsis), you could expect no more than 96.18 Mbps over IPv4.

Goodput = 100 * ( ( 1500 - 20 - 20 ) / ( 1500 + 18 ) )

I would run a wired speed test early in the morning when network was quite to get my baseline speed then set my cake parameters at 90% of the results. Then I discovered if I just set the speeds to what my test results were cake reduced the speed by about 10% on my next test. For an easy set and forget QoS, cake works very good.

 

[taffeys]

I'm using the Cake that comes with Merlin, no additional addons. It's recommended to define 95% of the actual speed but what I see is it is done automatically. I get 100/5 with QoS off. I set 100/5 on QoS settings. When I perform a speed test, I get 95/4.8 with Cake QoS on. Also, getting A+ on everything on the DSLReports test. Why is it recommended to set a 95% limit? Or do I still need to set 95% for best results?

I have been running Cake on my AX86U for some time. I have tried Cake at 95% and at Automatic and found the Automatic setting gives the best results on my network. I have a cable account with a 250/10 plan that routinely performs at 300/12, even with Cake enabled.

 

[Lynx]

If I understand correctly, and @dave14305 can correct me if I am wrong, the 'automatic' setting has no effect in CAKE. You can see whatever has actually been set by the GUI by issuing the command 'tc qdisc ls' in the terminal. I think 'automatic' is a little misleading there because it suggests that the bandwidth may be automatically adjusted, which would be cool, but is actually a huge challenge to implement well for those connection types that are variable. I wrote a bash script in OpenWrt that adjusts the CAKE bandwidth automatically, which I think could be made to work in Asus Merlin, but this is only needed for truly variable connections like LTE or Starlink, and its use would be counterproductive for fixed bandwidth connections since it would unnecessarily oscillate around the true bandwidth.

And consider this:

At 100Mbps with 1500 MTU, and overhead of 18 (docsis), you could expect no more than 96.18 Mbps over IPv4.

Goodput = 100 * ( ( 1500 - 20 - 20 ) / ( 1500 + 18 ) )

And for upload this gives 4.8:

5*( ( 1500 - 20 - 20 ) / ( 1500 + 18 )

So 96.18/4.8, which is pretty much what you are seeing @mutlucan.

If you are seeing no bufferbloat in tests with setting 100/5 then I think you are set and can leave things be.

 

[dave14305]

the 'automatic' setting has no effect in CAKE. You can see whatever has actually been set by the GUI by issuing the command 'tc qdisc ls' in the terminal.

For Cake, setting “Automatic” in the GUI results in no bandwidth limits (defaulting to unlimited).

 

[Lynx]

For Cake, setting “Automatic” in the GUI results in no bandwidth limits (defaulting to unlimited).

Ah! Thanks.

So @taffeys seems like you are just disabling CAKE and bufferbloat is not an issue for your connection.

 

[taffeys]

Ah! Thanks.

So @taffeys seems like you are just disabling CAKE and bufferbloat is not an issue for your connection.

Looks like you are correct. I just tested with and without any QOS enabled and bufferbloat is minimal. Must be the result of my ISP rebuilding and spliting the node I'm on earlier this month and increasing DL from 150 to 250 (300 actual) Mbps.

 

[Lynx]

Out of curiosity, what do you get on the waveform bufferbloat test? Could you post a link?

 

[taffeys]

Out of curiosity, what do you get on the waveform bufferbloat test? Could you post a link?

Bufferbloat Test by Waveform

 

[Lynx]

Hmm. I think in your position I would still want to use CAKE to prevent that level of bufferbloat. That is if your router can handle it at this rate?

 

[taffeys]

Hmm. I think in your position I would still want to use CAKE to prevent that level of bufferbloat. That is if your router can handle it at this rate?

Why wouldn't an AX86U be able to "handle it at this rate"? No one here does interactive gaming. I tested Cake again at actual rate settings and all I got is lower actual rates and no improvement in latency. I've got the hardware to build a dedicated router if this ASUS box can't do better. Interested in your opinion.

Bufferbloat and Internet Speed Test - Waveform

View the full results, and test your own bufferbloat

www.waveform.com

 

[Lynx]

Great, then why not use CAKE and keep the latency low despite saturation.

 

[taffeys]

Great, then why not use CAKE and keep the latency low despite saturation.

We don't saturate this connection by any means. I'll keep Cake active for a bit to see if there is any actual observed and measurable benefit. Thanks for your input, I appreciate it!

 

[dave14305]

We don't saturate this connection by any means. I'll keep Cake active for a bit to see if there is any actual observed and measurable benefit. Thanks for your input, I appreciate it!

My suggestion is to keep lowering the download rate until the CPU is not being saturated 100%, to ensure CAKE is controlling the bandwidth. When you get CPU-bound on a core, it’s hard to tell who’s in charge.

 

[Lynx]

I think the AX86U should be able to handle these rates? Otherwise dialing down a bit to ensure CPU is not saturated seems like a good idea.

Having consistently low latency is far more beneficial for the overall experience than a few extra megabits per second on a speed test or a slightly shorter download time.

@taffeys the point for me is why wouldn't you want to ensure bufferbloat is prevented when the connection is saturated?

What do you have to lose by applying CAKE to safeguard against this? After all, this kind of capability is why one uses something like an Asus router with Asus Merlin rather than the router supplied by an ISP.

The feature is there and ready to be used. So use it!

 

[taffeys]

Late last evening while there was minimal network activity I ran multiple Waveform bufferbloat tests while incrementally lowering the download and upload settings. I settled on 240.0/8.5 Mbps which results in 235/8 measured rates with little or no increases in latencies under load. I did notice that during the download test Core #1 went to 90% with no increase in latency. I will see how everything performs today and this evening. So, it does appear that CAKE is effective.

 

[Kingp1n]

Those using CAKE while gaming...which priority would u recommend? Should I stay with the diffserv3 (default) or try diffserv4?