Bufferbloat and QOS Asus AC66 Merlin

[Rich Brown]

Yes, that's a lot of latency, but I agree it would be hard to trade away half your download speed...

I see that OpenWrt/LEDE is not an option for your AC66. I wonder if RMerlin would consider incorporating fq_codel as an option in the builds to allow performance comparisons between the two sets of algorithms.

 

[RMerlin]

I see that OpenWrt/LEDE is not an option for your AC66. I wonder if RMerlin would consider incorporating fq_codel as an option in the builds to allow performance comparisons between the two sets of algorithms.

Not possible with the prehistoric kernel used by the RT-AC66U.

The RT-AC66U also doesn't support Adaptive QoS, so enabling QoS will disable NAT acceleration. That will impact performance based on what other rules/features you have enabled.

 

[mogulman]

Tried stock firmware also. QoS limits the speed to about half.

 

[Rich Brown]

Not possible with the prehistoric kernel used by the RT-AC66U.

Ouch. That's too bad.

The RT-AC66U also doesn't support Adaptive QoS, so enabling QoS will disable NAT acceleration. That will impact performance based on what other rules/features you have enabled.

I have always wondered about Adaptive QoS - what are its major features? Where does it operate in the networking stack? Thanks.

 

[RMerlin]

have always wondered about Adaptive QoS - what are its major features? Where does it operate in the networking stack? Thanks.

Traffic scheduling is still done through the same kernel "tc" subsystem. Main difference is, traffic classification is done with a closed source kernel module, based on DPI rulesets, while Traditional QoS uses iptables to mark packets based on user rulesets. That's why Adaptive is able to work even with NAT acceleration enabled, as packets aren't market by iptables but by the proprietary kernel module.

 

[jthomas]

You could try to use the 'pure' bandwidth limiter settings instead of QOS. Hardware acceleration remains enabled if using just that. I experimented with it for no particular reason (on a RT-N66U) and the BW limiter can improve buffer bloat much the same as QOS does if you limit to ~95%.

 

[mogulman]

Got an RT-AC3200 to test with. Updated to the newest Asus stock firmware, with Adaptive QoS turned on. 180/24 (my exact speeds)...

Bufferbloat is much better...

 

[L&LD]

Got an RT-AC3200 to test with. Updated to the newest Asus stock firmware, with Adaptive QoS turned on. 180/24 (my exact speeds)...

Bufferbloat is much better...

Bufferbloat score is better, but those ping times are almost 4x worse than what you had before in post 8.

Wonder what the RT-AC1900P with it's 75% faster processor will do here?

 

[mogulman]

Going to try the rt-ac1900p tomorrow

 

[mogulman]

RT-AC1900p, same exact settings...

Ping times are better, but maybe its time of day related

 

[sfx2000]

You folks need to stop bragging about your big pipes, lol...