CakeQOS CakeQoS-Merlin v2.0.0 alpha

[Wade Coxon]

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.

Hmm, not really at all. That doesn't seem to match what you'd expect from Cake on a ~100Mbps connection.
If your unshaped speed tests reliably hit 120Mbps, then you would expect to see Cake working well at around 100-110Mbps.

You would want to enable the "Filter duplicate ACKS" enabled on the upload, but other than that, the default Cake options should all be fine.

Your upload control seems ok since you are seeing +0ms there, though I would have expected about +5ms there too....maybe that is a clue to something odd happening?

What have you shaped your uploads to? That number should be about 5Mbps or less. I find that when your bandwidth limits are less than 10Mbps or so, you need to sacrifice a larger percentage of the total to QoS in order for it to work well.

What results do you get without any QoS?

 

[juniorsweet]

Hmm, not really at all. That doesn't seem to match what you'd expect from Cake on a ~100Mbps connection.
If your unshaped speed tests reliably hit 120Mbps, then you would expect to see Cake working well at around 100-110Mbps.

You would want to enable the "Filter duplicate ACKS" enabled on the upload, but other than that, the default Cake options should all be fine.

Your upload control seems ok since you are seeing +0ms there, though I would have expected about +5ms there too....maybe that is a clue to something odd happening?

What have you shaped your uploads to? That number should be about 5Mbps or less. I find that when your bandwidth limits are less than 10Mbps or so, you need to sacrifice a larger percentage of the total to QoS in order for it to work well.

What results do you get without any QoS?

Thanks for these suggestions. I've enabled the duplicate ACKS filter and have gotten good results (only about +20ms) with bandwidth limits of 100/5. Not sure why I had to go all the way down to 75 earlier to see improvements. I did try lowering the upload to 4 or 4.5, but I found that even at 5 I was still getting +0ms upload latency.

Without any QoS I'm seeing latency of +57ms down and +83ms up.

 

[Hazel]

Could someone take a look at my stats and see if there's room for improvement? I have a 300/30Mbit Docsis 3.0 Cable connection, have limited bandwidth to 90% on both download and upload. All settings are default expect for 'Filter Duplicate TCP Acks' on uploads, which I enabled, as per recommendation earlier in this thread. I noticed that in the evenings Netflix (4K/UHD subscription) has become some what laggy and I don't understand why, as the network is relatively quiet at those times (no gaming or YT watching kids etc) and I don't possess the knowledge to understand what I can do to further improve performance. I've disabled conmons' pingtest and spdMerlin's speedtest and AutoBW, to make sure they're not causing the lags.

Here are my stats. For full resolution please click this link to imgurl: Screenshot CakeQoS stats.

Updated stats on imgurl.

Thanks in advance for your guidance.

 

[Wade Coxon]

I noticed that in the evenings Netflix (4K/UHD subscription) has become some what laggy and I don't understand why, as the network is relatively quiet at those times (no gaming or YT watching kids etc) and I don't possess the knowledge to understand what I can do to further improve performance.

If it is happening in the evenings during "Netflix hour", then that likley coincides with your ISP's peak traffic period, since all their customers will be online and streaming their media simultaneously.

If you wait until that time of evening, and then turn off QoS, and do a speed test on your un-loaded Internet connection, you will likely find that your maximum speed will not reach 300 Mbps.

If that is the case, then QoS is not going to work well during the ISP's peak period, since they lack the bandwidth to reach your QoS's management limit. You are instead at the mercy of whatever QoS measures the ISP has in place (likely not great).

This will happen to varying degrees depending on which ISP you use, and how much they oversubscribe their own backbone connections.

Your options are to either set your QoS limit to a value that matches the speeds you see during evening peak periods, or switch to a ISP that can provide full bandwidth at all times of the day.

I specifically selected my ISP because they publish their backbone statistics, and how much bandwidth they provision, and they ensure that at all times, they can deliver full line speed.

 

[Hazel]

If you wait until that time of evening, and then turn off QoS, and do a speed test on your un-loaded Internet connection, you will likely find that your maximum speed will not reach 300 Mbps.

Thank you for your reply. In spdMerlins history I see that latency and jitter increases during peak hours, but promised speeds are easily achieved. Even during watching Netflix, speed measurements exceeds 300 Mbit (mostly vary between 306-310 Mb/s.)

Your options are to either set your QoS limit to a value that matches the speeds you see during evening peak periods, or switch to a ISP that can provide full bandwidth at all times of the day.

At 90% of max (theoretical) bandwidth (270/70) I was hoping to provide a enough to provide enough margin for QoS. Speedtests have never been below these thresholds. Unfortunately switching ISP's is not an option, due to several reasons. FTTH sucks here, and otherwise I'll have to switch to ADSL which doesn't provides these speeds at a similar price level and I'll lose all the benefits I know have combining TV, Internet, Interactive TV, free sports packages, major discount on cellular subscriptions with unlimited calls and data etc.

Are my stats looking okay? Should I adjust bandwidth up or downwards to optimize? Or is there anything else I can do?

Edit: I added another up to date screenshot with stats to my previous post.

 

[pdc]

Thank you for your reply. In spdMerlins history I see that latency and jitter increases during peak hours, but promised speeds are easily achieved. Even during watching Netflix, speed measurements exceeds 300 Mbit (mostly vary between 306-310 Mb/s.)

[...]

Are my stats looking okay? Should I adjust bandwidth up or downwards to optimize? Or is there anything else I can do?

Edit: I added another up to date screenshot with stats to my previous post.

I'm no expert, but your stats look very good, with hardly any drops (much less than 0.1%), so cake is not really active in moderating your data flow. Further adjustments are not likely to have any effect. I'm afraid I don't have any other ideas, but I think cake is not the problem here.

Netflix 4K/UHD only needs around 25 Mb/s so you have more than enough bandwidth to spare. Might there be processing running on whatever computer/device you are using to stream? Did you have other setups where Netflix was not laggy? Pushing 4K video is not always straightforward...

Edit: one other thought, you are close to the CPU limit for cake, you may want to watch your CPU usage during Netflix. I believe cake is single-threaded, so if one CPU is maxxed out that could impact your data flow.

 

[Hazel]

Edit: one other thought, you are close to the CPU limit for cake, you may want to watch your CPU usage during Netflix. I believe cake is single-threaded, so if one CPU is maxxed out that could impact your data flow.

Thanks, I'll keep an eye on that.

 

[Wistuplu]

I've disabled conmons' pingtest

I recommend you turn connmon ON. It uses close to zero bandwidth. It will give you insight like the packet loss on your line. For what it's worth, I just had my ISP working/testing in my home and working in my street. Some kind of problems are not under our direct control.

And make sure that you make your speedtests to a server "further away" than your ISP. Otherwise, they won't show issues at the border between your ISP and the rest of internet.

The fastest way to peace might be to choose a different quality level in Netflix ;-) I don't see much difference between 4K and FHD (but I don't have your eyes, your tele, or your room either....)

 

[Hazel]

I recommend you turn connmon ON. It uses close to zero bandwidth. It will give you insight like the packet loss on your line. For what it's worth, I just had my ISP working/testing in my home and working in my street. Some kind of problems are not under our direct control.

Connmon is on, however I enabled "Exclude ping tests from QoS" because otherwise my stats were being reset al the time because QoS is stopped and restarted.

And make sure that you make your speedtests to a server "further away" than your ISP. Otherwise, they won't show issues are the border between your ISP and the rest of internet.

I have, actually. A fixed speedtest server of another stable, reputable ISP.

The fastest way to peace might be to choose a different quality level in Netflix ;-) I don't see much difference between 4K and FHD (but I don't have your eyes, your tele, or your room either....)

Haha, my eyes are a disaster, but my room is small and the difference between 4K/UHD and the HD subscription is definitely visible... It's a joy to my eyes.

 

[Wistuplu]

EDIT: I didn't test yet but there is a chance that the stop/start can be achieved with
/tmp/qos start
and
/tmp/qos stop
.
See e.g. https://github.com/jackyaz/spdMerlin/blob/master/spdmerlin.sh#L1585
----------------------------------
Hello,

Thank you for cake_qos

The reboot of the router when turning Cake ON in Merlin 386.2 is a bit of a pain for me.

Will the old cake-qos (e.g. 1.0.8) work on 386.2 as it did on 386.1 ? Or is it now incompatible due to the built-in cake ?

Does the cake-qos 2.0 allow stopping/starting cake in CLI with the script, avoiding the reboot ?

Considering the read_me and the usage information for V2 https://github.com/ttgapers/cakeqos-merlin/tree/alpha#readme, I don't have much hope but I like dreaming ;-)

Take care
Best regards
W.


PS: On my AC86U, right after activating Cake in the QoS section of the webUI of 386.2, my log includes the following, related to the reboot;

May 6 20:53:20 rc_service: httpds 1217:notify_rc reboot
May 6 20:53:20 custom_script: Running /jffs/scripts/service-event (args: reboot)


Some references about the unexpected reboot:

Another question, and this is probably more applicable to main beta thread, bit I will ask here since there is a high concentration of cake users. If QoS is disabled, and I then enable QoS with cake, my router applies the setting and reboots. What's going on there, applying QoS shouldn't require that. Changing QoS settings and applying when QoS is already enabled doesn't reboot the router.

Any similar experiences?

I brought this up with Merlin and the code belongs to Asus and it’s very confusing what they’re doing on the QoS page.

 

[Morris]

Thank you for your reply. In spdMerlins history I see that latency and jitter increases during peak hours, but promised speeds are easily achieved. Even during watching Netflix, speed measurements exceeds 300 Mbit (mostly vary between 306-310 Mb/s.)



At 90% of max (theoretical) bandwidth (270/70) I was hoping to provide a enough to provide enough margin for QoS. Speedtests have never been below these thresholds. Unfortunately switching ISP's is not an option, due to several reasons. FTTH sucks here, and otherwise I'll have to switch to ADSL which doesn't provides these speeds at a similar price level and I'll lose all the benefits I know have combining TV, Internet, Interactive TV, free sports packages, major discount on cellular subscriptions with unlimited calls and data etc.

Are my stats looking okay? Should I adjust bandwidth up or downwards to optimize? Or is there anything else I can do?

Edit: I added another up to date screenshot with stats to my previous post.

Try enabling ACKs. Not sending ACKs for intermediate packets in the rotate window will require the entire rotate window to be retransmitted if the final ACK is missed. It also means that the next rotate window start may be delayed as it must wait for the final ACK.

Morris

 

[Hazel]

Try enabling ACKs. Not sending ACKs for intermediate packets in the rotate window will require the entire rotate window to be retransmitted if the final ACK is missed. It also means that the next rotate window start may be delayed as it must wait for the final ACK.

Morris

Thanks for your reply. How would I 'enable ACKS'? As mentioned before, I'm a total noob when it comes to QoS, so I don't have a clue what you mean. I've tried to read several Cake-related man pages, but they might as well be written in a foreign language, as I don't have sufficient knowledge on this topic. I know what QoS does in general, I think I have a (very) basic knowledge about how it does it, but that's where my knowledge ends.

The only setting I see with the word 'ACK' in the UI is Filter Duplicate TCP ACKs which I have enabled on Uploads. Seeing the explanation (Remove duplicate ACKs ... since only the last ACK is needed) that sounds like a good thing. Is that the setting you mean or do you mean something else?

Thanks for your understanding.
H.

 

[Wade Coxon]

The only setting I see with the word 'ACK' in the UI is Filter Duplicate TCP ACKs which I have enabled on Uploads.

Yea, thats the setting in question.

The suggestion is that your may want to re-test with that setting turned off again (enabling that setting results in far fewer TCP ACK packets being sent).

The setting is intended to save some of your upload bandwidth when downloads are hanging your connection. There are possible side effects to enabling the setting though.

 

[Wistuplu]

Here is a quote by a guru, which confirms the suggestion of Morris. In the quote, notice the 'if', the 'test' and the ratio.

https://www.reddit.com/r/HomeNetworking/comments/fvhr4w

if you want a little more bandwidth on your uplink, and your up/down bandwidth is at least 1x15 asymmetric, you can try the ack-filter option. It doesn't help on your downlink, nor on symmetric links.

 

[Wade Coxon]

Here is a quote by a guru, which confirms the suggestion of Morris. In the quote, notice the 'if', the 'test' and the ratio.

Morris was suggesting the opposite.

Hazel had already enabled the "Filter ACK" option on my suggestion because they have a 50:1 ratio on DL/UL.

Morris suggested to turn it off again in case the lagginess was being caused by an unreliable connection that was occasionally dropping those combined ACKs, thus requiring substantial retransmission delays.

 

[Wistuplu]

Hazel had already enabled the "Filter ACK" option on my suggestion because they have a 50:1 ratio on DL/UL.

Hello Wade, Morris,

My understanding is that Hazel's ratio is 10:1. See below.

I have a 300/30Mbit Docsis 3.0 Cable connection

Hence, I provided the reference and insisted on

In the quote, notice the 'if', the 'test' and the ratio.

The reference suggests to try "Filter ACK" if "your up/down bandwidth is at least 1x15 asymmetric"

So, in my understanding, Hazel's context is not where "Filter ACK" has a chance to be desirable.

Hence, in my eyes, the reference confirms the suggestion of Morris, i.e. to try without "Filter ACK".

I could have been more clear... I created my previous post on my smartphone and it didn't allow me to finetune the text as I would have liked...
Anyway, take care.
Warm regards
W.

 

[Hazel]

My understanding is that Hazel's ratio is 10:1

That's correct.

I have made a screenshot of the stats this afternoon, disabled the Ack filtering on uploads for now and will monitor data usage. Once the data usage is the same, I'll take another screenhot for comparison. As for the lag, I haven't experienced anything yesterday evening (while Filter duplicate ACKs was still on), so maybe there are other factors in play. Hopefully a comparison of the starts with comparable data usage will give some more clarity.

 

[Wade Coxon]

Hello Wade, Morris,

My understanding is that Hazel's ratio is 10:1. See below.



Hence, I provided the reference and insisted on


The reference suggests to try "Filter ACK" if "your up/down bandwidth is at least 1x15 asymmetric"

So, in my understanding, Hazel's context is not where "Filter ACK" has a chance to be desirable.

Hence, in my eyes, the reference confirms the suggestion of Morris, i.e. to try without "Filter ACK".

I could have been more clear... I created my previous post on my smartphone and it didn't allow me to finetune the text as I would have liked...
Anyway, take care.
Warm regards
W.

Ahh whoops, I had another post in my mind that was talking about a 6Mbit upload. Had the ratio calc overblown there

 

[Morris]

That's correct.

I have made a screenshot of the stats this afternoon, disabled the Ack filtering on uploads for now and will monitor data usage. Once the data usage is the same, I'll take another screenhot for comparison. As for the lag, I haven't experienced anything yesterday evening (while Filter duplicate ACKs was still on), so maybe there are other factors in play. Hopefully a comparison of the starts with comparable data usage will give some more clarity.

Is your Netflicks 4k streaming over WiFi or wired? If it is WiFi, what shows up for the connection under System Log / Wireless LOG. Most interested in RX & TX + RSSI. Also Streams. We all love the convivence of WiFi yet it is susceptible to multiple issues. A busy radio environment caused by a neighbor and be a huge problem.

 

[Hazel]

Is your Netflicks 4k streaming over WiFi or wired? If it is WiFi, what shows up for the connection under System Log / Wireless LOG. Most interested in RX & TX + RSSI. Also Streams. We all love the convivence of WiFi yet it is susceptible to multiple issues. A busy radio environment caused by a neighbor and be a huge problem.

Wired, every device in my household that has to stream (Smart TV, AppleTV's, set top box, PS4 Pro, cameras, my sons laptop) in one way or another is connected by ethernet. The only exception is my own laptop, which I barely use for streaming content. I usually watch Netflix 4K with the app from the TV itself, which is connected with a 1Gb connection (shielded CAT6) to a switch and from there to my router. There is no other streaming device in use at the same time at the same Gigabit switch. Total distance: less dan 10m / 30ft. According to the speedtests run on the TV, even during peak hours, the connection itself / the available bandwidth shouldn't be the issue here.