Need help with setting up CAKE SQM on RT-AX88U

[360NoScoper]

It’s been a while since I’ve posted. I upgraded to an ASUS RT-AX88U about 8 months ago. I installed Merlin onto it and a few months later, I upgraded to a 500Mbps FTTP connection along with upgrading all my Ethernet cables to a triple shielded version.

Since then, my network connection in online games has become a lot more stable. I rarely ever experience major lag.

I originally purchased this router to use the CAKE SQM that comes with Merlin. I thought I had set it up properly, but my connection still suffers from occasional bufferbloat. Both my download and upload latency can be quite variable. This is tested with my iPhone, which I know isn’t the best device to use as wireless connections will inherently experience latency issues, but I don’t own a PC - so I can’t test all these metrics using an Ethernet connection. The only device I use via Ethernet is my PS5 and that doesn’t have a web browser.

Can anyone explain to me the best method to set up CAKE on my router? I want the best bufferbloat score, with the lowest latency/jitter I can achieve whilst still having a good download/upload speed.

My connection details:

ISP - BT
Type - FTTP
Download Speed - 500Mbps, guaranteed by BT to never drop below 425Mbps
Upload Speed - 73Mbps, usually averages around 60Mbps

 

[dave14305]

Your router’s CPU probably can’t handle shaping 500Mbit. Start with 250 and test. Watch the router’s CPU idle utilization in top while you test. Increase the bandwidth in small increments so long as the results are good and a CPU core doesn’t max out.

 

[360NoScoper]

Your router’s CPU probably can’t handle shaping 500Mbit. Start with 250 and test. Watch the router’s CPU idle utilization in top while you test. Increase the bandwidth in small increments so long as the results are good and a CPU core doesn’t max out.

Thank you for the reply.

Can this be reliably gauged using my iPhone or do I need an Ethernet connect PC?

 

[Tech9]

RT-AX88U has the same CPU cores as RT-AC86U @dave14305 has. I had both to compare - safe limit is about 300Mbps. It can't do 500Mbps. You may not need QoS on 500Mbps ISP line @360NoScoper. Test with no QoS first and see how it goes. This may be the best setting for you.

 

[360NoScoper]

RT-AX88U has the same CPU cores as RT-AC86U @dave14305 has. I had both to compare - safe limit is about 300Mbps. It can't do 500Mbps. You may not need QoS on 500Mbps ISP line @360NoScoper. Test with no QoS first and see how it goes. This may be the best setting for you.

Thanks, I’ll give this a try!

Just to clarify, will using my iPhone impact results too much? Is this best only tested with an Ethernet connected PC?

 

[Tech9]

You can't have Cake QoS on 500Mbps line with no throughput limitation no matter what you test with.

 

[360NoScoper]

You can't have Cake QoS on 500Mbps line with no throughput limitation no matter what you test with.

It’s currently set to 450Mbps and that’s roughly the calculated speed my PS5 downloads games at.

 

[Tech9]

It’s currently set to 450Mbps

Your router can't do 450Mbps with NAT acceleration disabled and Cake regardless of your settings.

 

[360NoScoper]

Your router can't do 450Mbps with NAT acceleration disabled and Cake regardless of your settings.

Wait, do I need to disable NAT acceleration before I enable CAKE?

 

[dave14305]

Wait, do I need to disable NAT acceleration before I enable CAKE?

Yes, CAKE cannot see the traffic if NAT acceleration is enabled. That’s why the CPU becomes the bottleneck at higher speeds.

 

[Tech9]

do I need to disable NAT acceleration

Both Runner and Flow Cache turn off automatically when you enable Cake QoS. The router restarts in the process.

 

[360NoScoper]

Both Runner and Flow Cache turn off automatically when you enable Cake QoS. The router restarts in the process.

Okay I’m gonna need more help on this. It’s making me pull my hair out.

I’ve experimented by lowering my download speed down to to 200-ish range and bufferbloat latency spikes STILL have not improved. I don’t know if I have my WAN Packet Overhead set wrong or done something wrong.

 

[Tech9]

It’s making me pull my hair out.

Because online bufferbloat testing tools are not accurate and you're perhaps fighting with non-existing problem. I told you what you need to do.

 

[360NoScoper]

Because online bufferbloat testing tools are not accurate and you're perhaps fighting with non-existing problem. I told you what you need to do.

How is that I’ve watched people on YouTube achieve A* then?

Also, something cannot be proven to be objectively better without testing. So how exactly are we supposed to verify efficacy? Give me a test website or software I can use to configure this crap. I’m sick of tweaking numbers with zero visible improvement.

 

[Tech9]

Continue watching YouTube and trying then, no problem. What I know for sure is there is no bufferbloat on fiber line below line saturation speeds, most ISPs apply QoS already and your tools to control downlink are limited. You are going to hurt your entire network performance 95% of the time in attempts to fix something that possibly happens 5% of the time and you may not even notice it. The fact something is available in firmware doesn't mean you have to enable it and configure it. If you had 20Mbps xDSL line I would suggest trying some sort of QoS. For your 500Mbps fiber line it's mostly pointless.

 

[360NoScoper]

Continue watching YouTube and trying then, no problem. What I know for sure is there is no bufferbloat on fiber line below line saturation speeds, most ISPs apply QoS already and your tools to control downlink are limited. You are going to hurt your entire network performance 95% of the time in attempts to fix something that possibly happens 5% of the time and you may not even notice it. The fact something is available in firmware doesn't mean you have to enable it and configure it. If you had 20Mbps xDSL line I would suggest trying some sort of QoS. For your 500Mbps fiber line it's mostly pointless.

It’s not pointless at all since the whole point of me upgrading my home network was to improve my online gaming experience.

Claiming fibre connections are free from bufferbloat is objectively false. I don’t understand why you reply to threads without giving answers.

I’ve asked for the correct WAN Packet Overhead for my connection - no answer.
I’ve asked for the correct CAKE settings to use - no answer.
I’ve asked for reliable websites/software to test QOS values - no answer.

 

[Tech9]

Sorry, I have nothing more to add here. For gaming you need low latency and not high speed. Your testing (online tools at line saturation), conclusions (constant bufferbloat presence) and QoS choice (CPU limited and NAT acceleration incompatible Cake QoS) are all wrong. If you test over wireless and not with wired PC - even worse. What do you want me to help you with - going in a wrong direction faster?

 

[360NoScoper]

Sorry, I have nothing more to add here. For gaming you need low latency and not high speed. Your testing (online tools at line saturation), conclusions (constant bufferbloat presence) and QoS choice (CPU limited and NAT acceleration incompatible Cake QoS) are all wrong. If you test over wireless and not with wired PC - even worse. What do you want me to help you with - going in a wrong direction faster?

Well duh, that’s the point of a QOS or SQM. Keep latency low whilst the connection is loaded. Do I need to explain to you that 2 + 2 = 4?

ASUS Merlin comes with CAKE, it’s not a wrong choice. Everyone on here recommends CAKE, I choose CAKE and that’s now wrong?

Firstly, you said it yourself that my router doesn’t use NAT Acceleration so that’s not an issue. Secondly, I lowered my download down to even 5Mbps which completely avoids CPU limitations and STILL get bad bufferbloat latency.

It’s very clear to me you don’t really know what you’re talking about. Just scraps of knowledge learned over the years, jumbled together without any tangible connection.

Are there any actual experts on this forum or what?

 

[dave14305]

I don’t know if I have my WAN Packet Overhead set wrong

When in doubt, set overhead high (e.g. 48). It might reduce throughput for possibly improved latency. It might also be that from your location, you are prone to external latency that no local QoS/SQM can correct for.

 

[360NoScoper]

When in doubt, set overhead high (e.g. 48). It might reduce throughput for possibly improved latency. It might also be that from your location, you are prone to external latency that no local QoS/SQM can correct for.

Cheers. What about the MPU and Mode?