CakeQOS CakeQoS-Merlin v2.0.0 alpha

[randomName]

In the Download stats section, is there suppose to be a 'Packets Sent' and 'Sent' ?

 

[dave14305]

In the Download stats section, is there suppose to be a 'Packets Sent' and 'Sent' ?

Yes.

 

[randomName]

Yes.

Ok. I thought it would be something like Packets Received.

 

[dave14305]

Ok. I thought it would be something like Packets Received.

It was discussed briefly in post 231:

CakeQOS - CakeQoS-Merlin v2.0.0 alpha

Right, but there is the process of verifying the version that might need to be checked by a 384.19 user. And Dave mentions 1.0.x, implying there might be something later than 1.0.8 either available or in the works. I think your question has already been answered here...

www.snbforums.com

 

[pdc]

I have noticed for a while now that every time I visit the CakeQOS-Merlin tab, the stats show very small numbers, like they were just reset. For example, I went there just now and it shows 11MB downloaded.

I don't know if it matters, but I have a custom /jffs/scripts/qos-start script, which adds some iptables rules that @dave14305 posted in the 386-2 beta pages a while back, to ensure Teams/Zoom take advantage of the Voice tin on upload.

 

[Zavar]

Are you using spdMerlin? If so, my understanding is that it stops and restarts QOS every time it runs, so it would reset the Cake stats.

 

[dave14305]

I don't know if it matters, but I have a custom /jffs/scripts/qos-start script, which adds some iptables rules that @dave14305 posted in the 386-2 beta pages a while back, to ensure Teams/Zoom take advantage of the Voice tin on upload.

This has no impact on the issue you posted, but the Zoom rule in that post is incorrect. Zoom's ports are 8801:8810, not 8001:8010.

 

[Wade Coxon]

And also on that note, has there been any behind the scenes progress towards the application of a basic set of rules on the download side?

 

[pdc]

Are you using spdMerlin? If so, my understanding is that it stops and restarts QOS every time it runs, so it would reset the Cake stats.

I am not running spdMerlin, though I am running connmon. I don't think it resets QOS though? It seems more like the logic to load the stats is also somehow resetting them.

I'm not too worried about it, just wanted to mention since this is alpha...

@dave14305 thanks for the note about Zoom, I will update my script.

 

[Hazel]

I am not running spdMerlin, though I am running connmon. I don't think it resets QOS though? It seems more like the logic to load the stats is also somehow resetting them.

Now that's it's possible to exclude pings from QoS in the latest version of connmon, I'm wondering whether Cake isn't stopped (and the stats reset). If you have this option enabled (exclude ping from QoS) it might do exactly the same as spdMerlin does: stop CakeQoS until it is done pinging and restart CakeQoS afterwards and then your stats have been set back to zero. Maybe ask on the connmon thread?

Edit: I think here's your answer: https://github.com/jackyaz/connmon/...1538297b458970b8646ab21542ed1ffb4ef3798d6R966. Seems like QoS is being stopped when you have Exclude ping from QoS enabled.

 

[pdc]

Now that's it's possible to exclude pings from QoS in the latest version of connmon, [...]

Edit: I think here's your answer: https://github.com/jackyaz/connmon/...1538297b458970b8646ab21542ed1ffb4ef3798d6R966. Seems like QoS is being stopped when you have Exclude ping from QoS enabled.

Thanks for tracking that down! Sure enough, that was set. I actually wanted connmon to measure pings WITH cake, so I changed the setting. Now I have stats!

 

[pdc]

Okay, a question now that my stats are working

I just did a 6GB download and got a 10% drop rate. I got good download speeds, but I'm guessing the drops are cake throttling, and I should consider dropping my bandwidth a bit? It's currently at 97% (82.5/85) so probably on the high side... I'll drop it down to 94% (80/85) and see how that goes.

Here are the full stats:

Spoiler

Tin 0
Threshold Rate	84,480 Kbit
Target	1.5 ms
Interval	30 ms
Peak Delay	9.97 ms
Average Delay	1.5 ms
Sparse Delay	16 μs
Backlog	0 B
Packets Sent	4,558,266
Sent	6.18 GB
Hash Indirect Hits	36,476
Hash Misses	24,875
Hash Collisions	0
Drops	455,757
ECN Marked Packets	0
Dropped ACK Packets	0
Sparse Flows	1
Bulk Flows	1
Unresponsive Flows	0
Max Packet Length	1,514
Flow Quantum	1,514

I also have ack filtering on upload (connection is 85/6 or 14:1 so pretty asymmetrical) and it ended up dropping 10% of the acks. Very few packet drops on upload though so I think that's good.

 

[dave14305]

Okay, a question now that my stats are working

I just did a 6GB download and got a 10% drop rate. I got good download speeds, but I'm guessing the drops are cake throttling, and I should consider dropping my bandwidth a bit? It's currently at 97% (82.5/85) so probably on the high side... I'll drop it down to 94% (80/85) and see how that goes.

Here are the full stats:

Spoiler

​

Tin 0
Threshold Rate	84,480 Kbit
Target	1.5 ms
Interval	30 ms
Peak Delay	9.97 ms
Average Delay	1.5 ms
Sparse Delay	16 μs
Backlog	0 B
Packets Sent	4,558,266
Sent	6.18 GB
Hash Indirect Hits	36,476
Hash Misses	24,875
Hash Collisions	0
Drops	455,757
ECN Marked Packets	0
Dropped ACK Packets	0
Sparse Flows	1
Bulk Flows	1
Unresponsive Flows	0
Max Packet Length	1,514
Flow Quantum	1,514

I also have ack filtering on upload (connection is 85/6 or 14:1 so pretty asymmetrical) and it ended up dropping 10% of the acks. Very few packet drops on upload though so I think that's good.

Your target and interval look low. Do you get the same values running tc -s qdisc at the command line? Or have you changed the rtt?

 

[pdc]

Your target and interval look low. Do you get the same values running tc -s qdisc at the command line? Or have you changed the rtt?

Sorry, when I changed the threshold to 80 Mb/s it reset the stats. However, yes I am running "regional" (rtt 30ms). I'm in the midwest US and it did seem like most sites were in that ping range.

Next time I do a large download I'll check tc output against stats. Currently with ~900 MB downstream they match. 730,771 packets sent and only 15,089 drops, though that's for general web browsing and not a high bandwidth download, if that matters.

 

[dave14305]

Sorry, when I changed the threshold to 80 Mb/s it reset the stats. However, yes I am running "regional" (rtt 30ms). I'm in the midwest US and it did seem like most sites were in that ping range.

Next time I do a large download I'll check tc output against stats. Currently with ~900 MB downstream they match. 730,771 packets sent and only 15,089 drops, though that's for general web browsing and not a high bandwidth download, if that matters.

You are probably seeing drops when packets take longer than 1.5ms (target) to leave the queue. I would reset this back to default (internet) and observe your drops.

 

[juniorsweet]

Question about using Cake with Wireguard

First off, big thanks to the developers of this cake-qos code! I have been running this on my ac-86u for a week or so now and I really appreciate having the GUI for easy tweaks and additional insight into the stats.

I have noticed during bufferbloat tests (waveform.com) that my latency is significantly higher when running the test from a computer with an active Wireguard VPN connection.

Is this an expected outcome? Are there any cake settings that I can tweak to improve speed/latency when connected via a Wireguard connection? I should add that I have a standard DOCSIS (comcast) internet connection (120/6), and right now I have cake configured for DOCSIS.

I've pasted two bufferbloat test results below to show the significant latency difference between tests from the same device without Wireguard (top) and with Wireguard (bottom).

Wireguard not running

With Wireguard running

 

[Morris]

VPNs were invented for security and have considerable overhead. You should not expect good performance using a VPN

Morris

 

[Wade Coxon]

I've pasted two bufferbloat test results below to show the significant latency difference between tests from the same device without Wireguard (top) and with Wireguard (bottom).

You should also be getting better than a B rating on your non-Wireguard test too.

How much of a margin do you have to add to your download bandwidth before that download latency drops below +10ms?

 

[Kingp1n]

VPNs were invented for security and have considerable overhead. You should not expect good performance using a VPN

Morris

I have a full time VPN with FlexQoS and get A/B bufferbloat rating. Without Flex I get C/Ds. I wld like to see if I would get the same rating with CAKE.

Maybe this weekend I'll try that!

 

[juniorsweet]

You should also be getting better than a B rating on your non-Wireguard test too.

How much of a margin do you have to add to your download bandwidth before that download latency drops below +10ms?

Good point. I hadn't really played around with this before...just set my bandwidth limits to 85% of the nominal connection speed (120/6). Interestingly, I found that I needed to lower my download bandwidth all the way down to 75Mb/s in order to start seeing download latency close to +10ms.

Does this sound right? Maybe there are other settings I need to consider adjusting in order to achieve low latency with a higher download bandwidth.