I'm using a hap ac2 router with router os 7.14.3 and i can't solve the bufferbloat issue on my docsis 3.0 connection. Tried every possible simple queue setting over years using cake, fq_codel, pcq,sfq and even red queues. My ping reaches values of even 3000ms during bufferbloat using my mikrotik device. During download when sqm is enabled I'm reaching 50% cpu usage on hap ac2. I don't know if all the cores are used during the queue. I don't know if the ac2 has a too weak cpu to handle queues at 400-500 mbit connection speed. My connection is cut down by 20% from 500 to 400 mbit. Even a 20% cut won't help. Tried default values of queue types and dozens of custom settings from forums on the internet and even mikrotik forums , and nothing helped.

I'm not a network specialist. I'm just a user looser xD and a gamer.

Currently solved my bufferbloat issue with opnsense on x64 pc with fq_codel shaper. And I got good results with that but maybe I will go back to the ac2 as the main router if bufferbloat can be solved with hap ac2.

AC2 is good for 1 Gb/s. 400-500 Mb/s shoud be no problem for it.
Have you enabled fasttrack? https://wiki.mikrotik.com/wiki/Manual:IP/Fasttrack
I would stop playing with queues for a simple home use scenario.

Without queues i have gigantic bufferbloat. I cant enable fasttrack because if its enabled simple queues not working. Queues are cpu intensive.

When fasttrack is disabled on hAP ac2, then max throughput is severely limited. My experience with IPv6 (no fasttrack support) shows that hAP ac2 can do somewhere around 350Mbps (give or take). Processing queues adds to CPU workload. So I guess you'd have to drastically reduce queue throughput (to something like 100Mbps for starter) for router to actually able to do something about it intentionally (rather than slowing down everything due to CPU overload).

docsis 3.0 connection

What model of modem are you using? Check what chipset it uses. I have read mixed accounts of whether the puma 6 issues were actually solved or not. It might be worth trying a different modem.

And if you haven't explored this already-- it may also be a signaling issue between your isp and the modem. Among either the many cables or connectors in the path, and either within your home or the ISP side.

Its not modem,cable or something else i got in use opnsense router maded from pc. When im using opnsense there are good results. Better than on my hap ac2.

Of course, the x86 based router will perform much better - no surprise there - just like a truck will do a better job at pulling a boat than a car. Small, low power ARM A7 SoCs were never designed to be desktop replacements, and few if any ARM based consumer priced routers could handle gigabit speed with QoS until recently. Likewise, the small hAP ac2 wasn't designed for your use case. Every process you run on your router (like NAT and QoS) has to be done on the CPU, unless it can be handled by an ASIC. Use the hAP as an AP or a switch, and be done with it.

BTW, cable is notoriously bad for gaming and unless you have devices that starve your connection non-stop, no QoS strategy can fix issues past your gateway, like throttling on over provisioned links during peak hours.

You can put your outbound queue on ether1, and your inbound queue on bridge.

Then you should be able to fasttrack it.

You probably need to set the bucket size in the queue to 0.01 (or maybe less)
Also need to add no-mark and all other packet marks you use to these two queues.

I made two separate queues in queue tree with fasttrack. Im getting grade B or C bufferbloat test. On opnsense im getting A+ .

Im doing it wrong? I just enabled fasttrack from defconf.

So I did some "measurements" of bufferbloat. I did it using waveform web test. It was 3 quite distinct setups with very distinct results. What was in common for all tests was my router and ISP line (router is hAP ac2, running 7.14.3; ISP line is 1000/100 GPON). So here are results:

1. testing using a mediocre android tab via wifi and using chrome ... tab somehow prefers the 2.4GHz band which limits it to 72Mbps air interface symbol rate
   the result was a C
2. testing using a decent laptop running Windows 11, via 1Gbps ethernet and using Firefox
   the result was a straight A+
3. testing an underpowered linux (atom processor), via 1Gbps ethernet and using Firefox, connected via ssh (with X11 forwarding ... which made GUI sluggish, I don't know it that affects results or not)
   the result was a B

A note: throughputs, reported by web app, were similar in the last two tests (the wired), downlink latency however was much worse in the last test ...


So what gives? It does show that buffer bloat test is very susceptible to all sorts of variables and actual buffer bloat is only one of them. Seeing how miserably performed the tab via wifi (with quite slow wireless connection) shows that it's not always the router. Seeing the difference in the two wired tests (where infrastructure up to the testing host was the same) tells me that the test can be very misleading, the differences can't be explained by infrastructure.

@Siarap, I'm not saying that your hAP ac2 doesn't spoil the result, your own tests running opensense proves the opposite. But: as I already wrote, your router has to be able to process the traffic without too much added delay and at rates you mentioned the needed CPU power is likely higher than hAP ac2 can provide. As others already noted, if the bufferbloat grade (whatever it actually shows) is important for you, you'll have to use router with adequate power ... and hAP ac2 doesn't seem to be one (in my case, the same router showed A+ result but it may have much simpler config). If you'd like to use Mikrotik device, then there are quite a few much more powerful ones (with a higher price tag as well, you may have to make some compromise there), such as RB5009.

I decided to stay at opnsense x86 router. Cant do much with hap ac2. To low processing power. hap ac2 serves currently as access point connected to the opnsense router. And i need good bufferbloat grade because i am a gamer. Need low ping no matter what network load is. Thanks for replies.

I made two separate queues in queue tree with fasttrack. Im getting grade B or C bufferbloat test. On opnsense im getting A+ .

Im doing it wrong? I just enabled fasttrack from defconf.

Watch the connections tab in IP firewall window. it shows weather a connection ia fast tracked or not. then also system resource CPU, where you can view the CPU load on each core separately - and not just the overall average of all cores. Queues typically just stress a single CPU core.

And as already said: very low bucket size. Or else too much packets bypass and you won't gain a good bufferbloat rating.

See, it's not as trivial with ROS. But it could eventually work as well.

I made two separate queues in queue tree with fasttrack. Im getting grade B or C bufferbloat test. On opnsense im getting A+ .

Im doing it wrong? I just enabled fasttrack from defconf.

Watch the connections tab in IP firewall window. it shows weather a connection ia fast tracked or not. then also system resource CPU, where you can view the CPU load on each core separately - and not just the overall average of all cores. Queues typically just stress a single CPU core.

And as already said: very low bucket size. Or else too much packets bypass and you won't gain a good bufferbloat rating.

See, it's not as trivial with ROS. But it could eventually work as well.

3 of 4 cores was loaded at 100% with fasttracked queue tree. Core 0 0% usage. Bucket size 0.001. And very big bufferbloat as result (grade C). Im back on opnsense with grade A+.