Endgame Bufferbloat Results: RT-AX86U Pro + CAKE SQM (+0ms Active Latency)

[Fonpap]

I can guarantee someone with the same bandwidth and quality ISP line and the same model router running latest stock Asuswrt with NAT acceleration enabled by default will have better user experience 99% of the time due to the following:

- there is no way to fix bad ISP line on user's end
- the ISP already applies QoS on all shared residential lines
- no bandwidth 0.3-0.5Gbps restriction (≥50% loss in this case)
- no firmware features restriction (3004 vs 3006 in this case)
- no need to install custom firmware (full featured ASUS App?)
- no need to expose to vulnerabilities with Nov 2004 firmware
- no need to waste time with Waveform (or other similar) website
- no need to waste time in fixing what is not broken in first place
- they can freely use other available features in firmware (the "perfection" may disappear the moment other firmware feature starts actively using the same core, assumed 100% available all the time)

Thanks for sharing your work and experience.

 

[Fonpap]

I can guarantee someone with the same bandwidth and quality ISP line and the same model router running latest stock Asuswrt with NAT acceleration enabled by default will have better user experience 99% of the time due to the following:

- there is no way to fix bad ISP line on user's end
- the ISP already applies QoS on all shared residential lines
- no bandwidth 0.3-0.5Gbps restriction (≥50% loss in this case)
- no firmware features restriction (3004 vs 3006 in this case)
- no need to install custom firmware (full featured ASUS App?)
- no need to expose to vulnerabilities with Nov 2004 firmware
- no need to waste time with Waveform (or other similar) website
- no need to waste time in fixing what is not broken in first place
- they can freely use other available features in firmware (the "perfection" may disappear the moment other firmware feature starts actively using the same core, assumed 100% available all the time)

Thanks for sharing your work and experience.

I appreciate the feedback, but it seems there is a fundamental misunderstanding of what this setup aims to achieve. You are talking about 'User Experience' for the 99% (speedtests and app features), while I am talking about Performance Engineering for the 1%.

To address your points specifically:

1. NAT Acceleration vs. Bufferbloat: NAT acceleration (Runner/Flow Cache) is great for hitting Gigabit numbers on a speedtest, but it is a 'black box' that bypasses packet processing. In competitive gaming (Warzone), I don't need 1000Mbps; I need Zero Variance. By using CAKE with manual CPU affinity (Mask F) on Merlin 388.8_4, I have achieved a jitter of 0.3ms and a max spike of 8ms under 100% synthetic load. A stable 38ms ping is infinitely more valuable than an unstable Gigabit line that spikes the moment someone else on the LAN start a 4K stream.
2. Security & Firmware (3004 vs 3006): Chasing a version number (3006) doesn’t automatically mean better security. Asuswrt-Merlin 388.8_4 is a highly mature, hardened build. RMerlin consistently backports CVE patches and binary blobs from the 3006 branch. In the Pro-gaming community, the consensus is: 'Better a hardened 388 than a buggy, bleeding-edge 3006.' Moving to 3006 right now means dealing with early-adopter bugs and losing the granular JFFS script control (smp_affinity, RPS pinning) that makes this 'Lab Perfect' stability possible.
3. ISP QoS: Residential ISP QoS is designed to protect the provider's backbone, not the user's micro-latency. If you rely on the ISP to manage your bufferbloat, you've already lost the battle at the first hop.

This is a 'Purist' build. I’ve traded raw, unusable speed for surgical precision. It’s a conscious choice to prioritize Latency Consistency over raw throughput. My 0.290ms stddev speaks for itself

 

[Viktor Jaep]

I appreciate the feedback, but it seems there is a fundamental misunderstanding of what this setup aims to achieve. You are talking about 'User Experience' for the 99% (speedtests and app features), while I am talking about Performance Engineering for the 1%.

To address your points specifically:

1. NAT Acceleration vs. Bufferbloat: NAT acceleration (Runner/Flow Cache) is great for hitting Gigabit numbers on a speedtest, but it is a 'black box' that bypasses packet processing. In competitive gaming (Warzone), I don't need 1000Mbps; I need Zero Variance. By using CAKE with manual CPU affinity (Mask F) on Merlin 388.8_4, I have achieved a jitter of 0.3ms and a max spike of 8ms under 100% synthetic load. A stable 38ms ping is infinitely more valuable than an unstable Gigabit line that spikes the moment someone else on the LAN start a 4K stream.
2. Security & Firmware (3004 vs 3006): Chasing a version number (3006) doesn’t automatically mean better security. Asuswrt-Merlin 388.8_4 is a highly mature, hardened build. RMerlin consistently backports CVE patches and binary blobs from the 3006 branch. In the Pro-gaming community, the consensus is: 'Better a hardened 388 than a buggy, bleeding-edge 3006.' Moving to 3006 right now means dealing with early-adopter bugs and losing the granular JFFS script control (smp_affinity, RPS pinning) that makes this 'Lab Perfect' stability possible.
3. ISP QoS: Residential ISP QoS is designed to protect the provider's backbone, not the user's micro-latency. If you rely on the ISP to manage your bufferbloat, you've already lost the battle at the first hop.

This is a 'Purist' build. I’ve traded raw, unusable speed for surgical precision. It’s a conscious choice to prioritize Latency Consistency over raw throughput. My 0.290ms stddev speaks for itself

Quick question... do you know how to respond without the use of AI? Like a normal person?

 

[podkaracz1996]

Why cant they incorporate the cake qos into the stock firmware i have no idea maybe @RMerlin can talk with them to push it?

 

[Viktor Jaep]

Why cant they incorporate the cake qos into the stock firmware i have no idea maybe @RMerlin can talk with them to push it?

I'm sure @dave14305 can give you a dissertation about that.

 

[Fonpap]

Quick question... do you know how to respond without the use of AI? Like a normal person?

Look, English isn't my first language, so yeah, I use tools to make sure my points come across clearly. I’d rather be precise than messy.

But the settings are mine, the testing is mine, and the 0.290ms stddev is sitting right there on my router. Whether I use a tool to structure the post or not doesn't change the CAKE results or the smp_affinity fix.

If you have a technical counter-argument about the bufferbloat or the RTT tuning, I’m all ears. That’s what we’re here for, right?"

 

[Viktor Jaep]

Look, English isn't my first language, so yeah, I use tools to make sure my points come across clearly. I’d rather be precise than messy.

Good for you! Great use case. It's just that when I see an AI response scraped directly off some tool and pasted into here, it rubs me the wrong way, makes my neck hair stand on end, and irritates my OCD greatly.

But the settings are mine, the testing is mine, and the 0.290ms stddev is sitting right there on my router. Whether I use a tool to structure the post or not doesn't change the CAKE results or the smp_affinity fix.

If you have a technical counter-argument about the bufferbloat or the RTT tuning, I’m all ears. That’s what we’re here for, right?"

Sadly, I see no need to deal with bufferbloat. I'm pretty happy with my vanilla results. Zero complaints, and have zero interest to obsessively drive these numbers down to something lower so I can get an A+.

 

[Fonpap]

Good for you! Great use case. It's just that when I see an AI response scraped directly off some tool and pasted into here, it rubs me the wrong way, makes my neck hair stand on end, and irritates my OCD greatly.


Sadly, I see no need to deal with bufferbloat. I'm pretty happy with my vanilla results. Zero complaints, and have zero interest to obsessively drive these numbers down to something lower so I can get an A+.

View attachment 71015

I totally get the OCD part, no hard feelings! To be honest, I'm just as obsessed with my network tuning as you are with 'clean' posts.

Looking at your results, they are definitely great for 'vanilla' use. But that +26ms on Download Active is exactly what I'm fighting against. In a competitive shooter like Warzone, that’s the difference between a hit and a miss when the line is under load.

For me, the 'A' grade and that warning sign on 'Low Latency Gaming' in your screenshot is the reason I spent time on CAKE and smp_affinity. We just have different goals: you're happy with 'good enough,' and I’m chasing 'Lab Perfect.'

Enjoy your setup, and I’ll see you in the lobbies (hopefully with my 0ms jitter advantage!

 

[dave14305]

the reason I spent time on CAKE and smp_affinity

Does this smp_affinity only work for pppoe?

 

[Tech9]

Looking at your results, they are definitely great for 'vanilla' use. But that +26ms on Download Active is exactly what I'm fighting against. In a competitive shooter like Warzone, that’s the difference between a hit and a miss when the line is under load.

What is a “line under load” when gaming? I guess, your game needs under 50Mbps, no?

 

[Fonpap]

Does this smp_affinity only work for pppoe?

Not at all, it works for any interface! However, it’s specifically critical for PPPoE connections because PPPoE is single-threaded in Linux. Without manual steering, the router often dumps all the heavy lifting (softirqs) onto Core 0, creating a bottleneck even if the other cores are idle.

By using smp_affinity (and RPS for the Ethernet interfaces), I'm forcing the OS to distribute the interrupt load. On my RT-AX86U Pro, moving the ppp0 interrupts to a different core mask significantly reduced micro-jitter and prevented those 'hidden' spikes during heavy gaming sessions.

It's basically about ensuring that the CPU's 'scheduling' doesn't interfere with the packet's path.

 

[dave14305]

Not at all, it works for any interface

What are the interrupts for the non-ppp0 interfaces?

 

[Tech9]

RMerlin consistently backports CVE patches and binary blobs from the 3006 branch.

This is incorrect, by the way. There are no continued development on two separate firmware branches for RT-AX86U Pro model. The last 3004 base firmware for it contains security fixes before Nov 2024. Since then ASUS released a few urgent updates. Right now another one is happening because of discovered critical vulnerability. Some folks lost their routers in last 12 months to malware damaging factory radio configuration.

I guess, your game needs under 50Mbps, no?

I'm not a gamer, but this is what I find for connection requirements:
- Download Speed: 50-100 Mbps ensures smooth, uninterrupted play, especially if others are using the network.
- Upload Speed: A consistent upload speed of at least 5 Mbps is recommended for stable gameplay.
- Latency: Ideal (0-30ms): Near-instant responsiveness, minimal lag; Good/Comfortable (30-60ms): Consistent, standard experience; Noticeable Lag (60-100ms): Observable delays, competitive disadvantage.
- Data Usage: Warzone typically uses roughly 160MB to 500MB of data per hour, which is relatively low compared to the bandwidth available on most broadband plans.

Now... @Fonpap, you have to show your gaming profile with great gameplay achievements in order to back up the practical aspect of your optimization efforts. If you really kick @ss in Warzone as a result of personal gaming skills and specific setup advantages - respect. If you don't... you're wasting time.