p/fix-bufferbloat-on-mikrotik-without-disabling-fasttrack/

[giscus[bot]]

p/fix-bufferbloat-on-mikrotik-without-disabling-fasttrack/

Enable/Setup FQ-CoDel or CAKE SQM on RouterOS using Queue Tree

https://www.animmouse.com/p/fix-bufferbloat-on-mikrotik-without-disabling-fasttrack/

[dtaht]

This was really good, thank you! Two quibbles. CAKE does more, and the results won't show up on the waveform test. I described that over here as best as I can describe: https://www.reddit.com/r/eero/comments/u7xm83/gen_2_sqm_vs_gen_3_sqm_stick_with_gen_2_if_you/

However, your result for cake being slightly slower than fq_codel might actually be correct and not measurement noise. Throwing flent at it ( https://blog.cerowrt.org/post/flaws_in_flent/ ) would also verify if you are, indeed correct in being able to only slice off that tiny percentage of the bandwidth. or not.

Was this a fiber connection?

[AnimMouse]

This was really good, thank you!

You're welcome, thank you also for helping in the fight against bufferbloat!

CAKE does more, and the results won't show up on the waveform test. I described that over here as best as I can describe

I read about that CAKE is more modern than FQ-CoDel, I tried both algorithms and FQ-CoDel won according to Waveform bufferbloat test.

However, your result for cake being slightly slower than fq_codel might actually be correct and not measurement noise. Throwing flent at it ( https://blog.cerowrt.org/post/flaws_in_flent/ ) would also verify if you are, indeed correct in being able to only slice off that tiny percentage of the bandwidth. or not.

Good idea, will try using Flent to determine if CAKE is really better than FQ-CoDel.

Was this a fiber connection?

Yes, I'm using Converge ICT 250 Mbps fiber connection.

[GreenBattery]

Cake appears to be really good, up to a bandwidth of about 450mbps on my mikrotik hap ax3, which has a quad-core ARM cpu that can boost up to 1.8ghz. Beyond 450mbps (I've tested 500mbps and higher), some CPU limitation kicks in, and the results of using cake become much worse than FQ-Codel in terms of latency and maximum throughput, not just as measured in tests, but also in online gameplay experience. Probably this is what @AnimMouse experienced when comparing CAKE and FQCodel.

On my 1G/100M connection, any sort of limit at 450Mbps on the downstream is effectively an A+ grade on buffer-bloat tests and perhaps the impact of any AQM is minimal when throughput is limited to 50% of available bandwidth.

This means that FQCodel is the only choice available on my hardware... and it is very hard to find reference data or benchmarks on what sort of CPU power one needs to shape traffic using Cake for commonly available home internet speeds.

[wpsaputra]

Nice article, I'm using fq_codel with fasttrack enable and default mikrotik setting, make my hex poe keep running under 10% cpu! With fasttrack disable it can spike up to 50% during bufferbloat test.
I have a question, did your queue tree settings affect lan speed connection between pc on default bridge interface (not yet tested it)? Anyway to disable fqcodel between host on 192.168.88.0/24 network? thanks

[AnimMouse]

I have a question, did your queue tree settings affect lan speed connection between pc on default bridge interface (not yet tested it)?

Nope, this queue tree setting does not affect LAN speed connection between PC on the default bridge interface. It only affects the connection from your router going to the internet and from the internet going to your router.

[gabriele-v]

What if I have multiple LAN interfaces?

[AnimMouse]

Either bridge them all and apply the queue tree to that bridge, or create a queue tree for each of the interface just like this.