FQ-Codel not working well in practice - broken implementation ?

[Find the Door]

I use the freshjr qos script slightly modified to allow the manual bw setting to be manipulated by an automatic speedtest I run on a pi.

With this combo I get all A+ or A rated from dslreports, 0 set as overhead and up and down speeds set to 98% of what the bandwidth test gets.

I am also running fq-Codel

How do you modify that? Is that what I'm missing that's leading to fq_codel doing effectively zilch for me?

 

[Vexira]

How do you modify that? Is that what I'm missing that's leading to fq_codel doing effectively zilch for me?

The fq_codel in the Merlin firmware is not the full feature set if you did some research, I asked Merlin about this and cake already, problem is that some units are running an older kernel version and the other is adaptive qos is closed source, also have you set you permissions packet overhead correctly and applied the expir commands to lower total bandwidth for fq_codel to work properly?
Also if you want I can see if I can't help you get it working better.

 

[Find the Door]

The fq_codel in the Merlin firmware is not the full feature set if you did some research, I asked Merlin about this and cake already, problem is that some units are running an older kernel version and the other is adaptive qos is closed source, also have you set you permissions packet overhead correctly and applied the expir commands to lower total bandwidth for fq_codel to work properly?
Also if you want I can see if I can't help you get it working better.

Yeah definitely that'd be great if you have the time.

 

[Vexira]

Just pm me

Yeah definitely that'd be great if you have the time.

 

[FreshJR]

Has anyone else test FQ-Codel against SFQ ?

I get a significantly faster RAMP to max speed (a signature trait of CoDel) compared to sfq.
My bufferbloat is +5-10ms above average.

These two things imply to me that it is working as expected.

The only weird thing is that a have a 1sec burst of bufferbloat for the first second when the dslreports speedtest is starting (other users have reported that it doesn't happen to them).

I will post gif results of before/after.

Edit:

Spoiler: bandwidth

Spoiler: download bloat

Spoiler: upload bloat

 

[Find the Door]

I get a significantly faster RAMP to max speed (a signature trait of CoDel) compared to sfq.
My bufferbloat is +5-10ms above average.

These two things imply to me that it is working as expected.

The only weird thing is that a have a 1sec burst of bufferbloat for the first second when the dslreports speedtest is starting (other users have reported that it doesn't happen to them).

I will post gif results of before/after.

Edit:

Spoiler: bandwidth

Spoiler: download bloat

Spoiler: upload bloat

I don't know - I personally tried out Traditional QoS last night and had a pleasant experience with it (a few more tweaks needed).

Why does everyone say it's broken or not complete?

 

[RMerlin]

I don't know - I personally tried out Traditional QoS last night and had a pleasant experience with it (a few more tweaks needed).

Why does everyone say it's broken or not complete?

Depends on your router model. On some models it's completely broken.

I got a potential fix for 384.5, unsure yet however if it will address all its issues or only some of them.

 

[Find the Door]

Depends on your router model. On some models it's completely broken.

I got a potential fix for 384.5, unsure yet however if it will address all its issues or only some of them.

I am running the Alpha right now. Possibly that's why it's working for me. I have an AC86U.

Curiously I only had one instance last night of an upload spike but aside from that it seems to perform pretty well after adjusting the upload and download thresholds.


I noticed I did have to limit it on the download and upload side as far as maximum usage aside from priority itself. My connection was infinitely more stable though for the time being.

 

[FreshJR]

I don't know - I personally tried out Traditional QoS last night and had a pleasant experience with it (a few more tweaks needed).

Why does everyone say it's broken or not complete?

It was just recently fixed (last month) by a member on this forum.

I haven't checked how ASUS's traditional QOS structure is setup, but now that I learned significantly more how QOS works in reality I can see a traditional QOS solution outperforming the Asus/TrendMicro implementation.

I think ideally setting up a root HTB and then on the next level down each LAN client would be a child of that HTB with Highest/High/Med/Default/Low per device configurable at this level. Within each child, the traffic should then be prioritized again into High, Med, Low according to traditional rules.

This implementation would probably give BETTER results than Asus/TrendMicro AdaptiveQOS, as first they sort by traffic type but in the next level down device priority is broken which can cause fights over bandwidth.

As for the bufferbloat, I have no idea why this is happening to some of you guys. !

 

[Find the Door]

It was just recently fixed (last month) by a member on this forum.

I haven't checked how ASUS's traditional QOS structure is setup, but now that I learned significantly more how QOS works in reality I can see a traditional QOS solution outperforming the Asus/TrendMicro implementation.

I think ideally setting up a root HTB and then on the next level down each LAN client would be a child of that HTB with Highest/High/Med/Default/Low per device configurable at this level. Within each child, the traffic should then be prioritized again into High, Med, Low according to traditional rules.

This implementation would probably give BETTER results than Asus/TrendMicro AdaptiveQOS, as first they sort by traffic type but in the next level down device priority is broken.

Yeah it's giving me great results in regards to Bufferbloat. I sort priority by way of Mac Address. Seems to be functioning so far.

 

[Find the Door]

Depends on your router model. On some models it's completely broken.

I got a potential fix for 384.5, unsure yet however if it will address all its issues or only some of them.

Also when can we expect Alpha 2 that will include that fix?

 

[RMerlin]

I am running the Alpha right now. Possibly that's why it's working for me. I have an AC86U.

No, the alpha builds currently in the Test Builds folder do not contain the patch yet. I only merged and compiled it tonight, I haven't uploaded anything yet.

 

[RMerlin]

Also when can we expect Alpha 2 that will include that fix?

Don't know yet. Working on a lot of things at the same time, so it will depend on when I feel I have something that's usable enough to upload.

 

[Find the Door]

No, the alpha builds currently in the Test Builds folder do not contain the patch yet. I only merged and compiled it tonight, I haven't uploaded anything yet.

Don't know yet. Working on a lot of things at the same time, so it will depend on when I feel I have something that's usable enough to upload.

Roger that. I'll be on the lookout - what exactly is wrong with traditional at the moment?

 

[RMerlin]

what exactly is wrong with traditional at the moment?

A lot of things, including that the traffic classifier isn't properly handling traffic. One of the visual symptoms is it looks as if the upload and download limits are reversed on the webui - except they aren't.

 

[Find the Door]

A lot of things, including that the traffic classifier isn't properly handling traffic. One of the visual symptoms is it looks as if the upload and download limits are reversed on the webui - except they aren't.

Gotcha - yeah I experienced quite a few issues with persistent random spikes last night trying to use fq_codel. Happened whether or not I was using Traditional or Adaptive.

Something is just not right with that particular queing discipline.

 

[Gar]

What packet size and protocol for fiber? 100mb/s.

sfq vs. fq-codel still a matter of individual setup or an obvious choice in Merlin now?

 

[skeal]

What packet size and protocol for fiber? 100mb/s.

Not sure what you mean by protocol but I have fiber and use 4 as an overhead setting, fiber is converted to ethernet by an ONT.

 

[Gar]

Not sure what you mean by protocol but I have fiber and use 4 as an overhead setting, fiber is converted to ethernet by an ONT.

oops...wan packet preset...not protocol.

 

[skeal]

oops...wan packet preset...not protocol.

Wan packet preset would be "Ethernet VLAN" in your case. Same as myself.