CakeQOS CakeQOS-Merlin

[ugandy]

I hear you - but genuinely believe that Flexqos, once it matures, will be the best way to handle shaping as it is incredibly capable of being customised. Its predecessor worked really well for me - despite the fact that I don't have a need for multi-level shaping or granular configuration into 8+ categories of traffic.

i gave FlexQos a try, it treated traffic according to the priorities, prevented packet loss, prevented bufferbloat when i saturated the download, but it didn't prevent bufferbloat when my upload line was saturated

 

[ugandy]

found this online about ack filtering:


> What is ack-filtering? How is it important? What's the difference
> between ack-filter-aggressive and ack-filter?
Simply put, we sometimes found that downstream throughput was limited by acks on the reverse path, on highly asymmetric links (such as ADSL) loaded in both directions simultaneously. Because acks are themselves very small, AQM activity by itself wasn't enough to clear the backlog.

Ack-filtering looks for consecutive acks accumulating in a flow queue, and deletes the ones which are not needed, improving throughput and efficiency. The "aggressive" version deletes more acks, but I think we found it hit too many under certain conditions.

It's harmless to leave it switched on in the non-aggressive mode. If you're a purist who doesn't like dropping packets, then you can switch it off.


https://bloat.bufferbloat.narkive.com/PCQIrEs7/benefits-of-ack-filtering

 

[skeal]

New hotfix settled the reboot problem on my AX88U, thank you @Adamm

 

[ttgapers]

I tried autorate-ingress in a quick test before. maybe i set it up incorrectly, but in my test, the speed limits kept swinging wildly and way below the bandwidth available for both DL and UL. i'll try again. my comcast UL swings between 20Mbps to 3Mbps during the day, so autorate sounds like something that would be needed in my config. i'll do another test

I assume you confirmed the setting was enabled

Thanks for testing!

 

[ugandy]

I assume you confirmed the setting was enabled

Thanks for testing!

it was.

i just repeated the test.

the limits do appear to constantly change (via cake-qos status) and quite lower that what is available, at least for DL.
dslreports gave me a B on bufferbloat (i never get more than A even with cake).

but, using fast.com, speedtest,net, etc:

i see (pingplotter) that my latency is kept always around ~15ms, both during the upload test and the download test. sounds good to me!

my max download was 310Mbps without autorate and 280Mbps with autorate. not too bad!
UL speed was way better with autorate, than when i set a hard limit of 4Mbps (to deal with comcast UL swinging between 5-20 during the day). with autorate i can use the full UL bandwidth available.


i need to do a 24hr test and see how bandwidth shapes out in spdmerlin. will report back tomorrow.

 

[rgnldo]

Thanks @rgnldo

Tracking here: https://github.com/ttgapers/cakeqos-merlin/issues/42

With the autorate-ingress option, it is not necessary to inform DL / UP rates. It is a solution for environments with bandwidth fluctuation, prioritizing navigation, reducing latency and congestion. Cake is excellent for this situation.

 

[jwortiz]

So my two cents on this @Adamm would be to keep key functionality inline with previous modules/addons. So, if "disable" functionality "disables" until "next reboot" and/or "enable", then I am ok to open the issue and get this parked.

On the reboot item, I am also open to whether this is an issue, to log it to github/issue tracker so a dev(s) can track it.

Hope that helps.

I feel the reboot item is important for the folks that schedule their router reboots (like me). I have it rebooting once a week. Thanks!

 

[brummygit]

I'd also decided to give autorate-ingress a try as I'm having issues with my primary line at the moment and sometimes failover to 4G using dual WAN. Automatic rate setting when this happens would be fantastic. I configured it leaving my current bandwidth settings in place as the documentation says you can specify them as a starting point.

The speed adjustment happens quickly (you can see it in my DSLReports test below) and falls back to lowers figures quickly as the traffic reduces. Bufferbloat isn't great on the uplink but probably better than without cake or with incorrect speed settings.

 

[lilstone87]

This is gonna sound like a noob question.. and I don't mind at all. However I am wondering, is there anything I should be setting under "extra options"? This is new to me. I have docsis based internet, if that matters. I hate asking this, as it might come off silly to some of you. However I just don't know what I should be setting, or using for this section of options.

 

[ugandy]

This is gonna sound like a noob question.. and I don't mind at all. However I am wondering, is there anything I should be setting under "extra options"? This is new to me. I have docsis based internet, if that matters. I hate asking this, as it might come off silly to some of you. However I just don't know what I should be setting, or using for this section of options.

try "docsis ack-filter autorate-ingress"

 

[lilstone87]

try "docsis ack-filter autorate-ingress"

Will do, and thank you.

 

[ttgapers]

With the autorate-ingress option, it is not necessary to inform DL / UP rates. It is a solution for environments with bandwidth fluctuation, prioritizing navigation, reducing latency and congestion. Cake is excellent for this situation.

Exactly!! I think @ugandy is testing.

 

[brummygit]

Exactly!! I think @ugandy is testing.

My experience is that if you can set your speeds to known values it will perform better than autorate-ingress so you could also try droping that parameter off and see how it compares. With the new menu system it's very easy to add and remove parameters.

 

[lilstone87]

Well I set my limits to 370mbps down, and 28mbps up. Ran a test, results were pretty damn good.

 

[L&LD]

@lilstone87 looks good! You may need to update your signature now?

 

[ugandy]

My experience is that if you can set your speeds to known values it will perform better than autorate-ingress so you could also try droping that parameter off and see how it compares. With the new menu system it's very easy to add and remove parameters.

fixed limits will work better if you don't have comcast, where my UL line goes from 20Mbps to 2Mbps during the workday

 

[ugandy]

some numbers.
my line is 600/15 and i normally max at 550/17 at night, without cake, but bufferbloat is horrible.

if i set cake with fixed limits, I can get as high as 330Mbps downloads; my UL is a problem though, since comcast UL bandwidth varies throughout the day (2Mbps-20Mbps)

with autorate-ingress, i'm getting

where my DL max is 286Mbps, but my UL speed is, so far, nicely tracking my UL available bandwidth (you can see it going down to ~5Mbps at 9:42am).

latency averages 10ms-20ms (ping plotter), except for a few seconds during the initial DL/UL burst, where it can reach !100ms, while cake adapts (normal?).

i think my plan going forward, since I'm keeping comcast, is to downgrade my service to 300/10 and use autorate-ingress.

my contracted line: 600/15 dslreports bufferbloat D
my cake setting: 300Mbps 10Mbps "docsis ack-filter-agressive autorate-ingress"

note1: cake-qos status will show ever changing values for limits, and lower than the bandwidth available, when auto rate is on, so i stopped looking at it, and i'm going only by spdMerlin and pingplotter
note2: home network traffic was very light during this period.

 

[lilstone87]

@lilstone87 looks good! You may need to update your signature now?

I did, removed the 2nd part. As I will test between this, and FlexQoS. As I do like the controls FlexQoS offers, but I know Cake>FQ_Codel. Maybe one day, it will be possible to have more control, plus have Cake as an option.

Overall.. I am very happy to see the recent work being put into this Cake project. Plus the work that's being put into FlexQoS. As it's nice someone is picking up on what FreshJR left behind. As Asus standard QoS system isn't very good, as most of us know already. Anyways goodjob to everyone helping out on both projects recently, just know a lot of people appreciate it!

 

[L&LD]

On the initial release of this script, I did my testing here.

https://www.snbforums.com/threads/release-cakeqos-merlin.64800/#post-595227

With v1.0 of this script, nothing has changed with my 1Gbps symmetrical Fibre connection. Still getting up to 480Mbps down and 520Mbps upload with up/down inputs of 1000/1000 in the script and besteffort selected.

I also tested with autorate-ingress and found the speeds to plummet to 320Mbps down and about 380Mbps upload.

DSLreports showed A for all tests (without the script, A+ for all).

Fast.com showed 1ms, 5ms unloaded, loaded latencies without the script and 2ms, 18ms unloaded, loaded latencies with the script used. With autorate-ingress, the loaded latencies increased to over 24ms.

A final test I did with setting cake-qos download/upload speeds, both to 250Mbps, showed speeds of 240Mbps up/down and loaded and unloaded latencies of 2ms each.

I would agree with @ugandy that where speeds vary greatly, autorate-ingress will be the best choice overall (at least for cable ISPs), but the highest speeds possible is manually setting them for the current ISP conditions.

HTH.

Great job on how fast this has come together for all concerned!

If we can have a list to choose from for the 'Extra Options' (I don't know how else we would know about them, except by reading the thread otherwise), it would be the icing on the cake!

This seems ready for inclusion in amtm any moment now.

 

[ugandy]

I did, removed the 2nd part. As I will test between this, and FlexQoS. As I do like the controls FlexQoS offers, but I know Cake>FQ_Codel. Maybe one day, it will be possible to have more control, plus have Cake as an option.

Overall.. I am very happy to see the recent work being put into this Cake project. Plus the work that's being put into FlexQoS. As it's nice someone is picking up on what FreshJR left behind. As Asus standard QoS system isn't very good, as most of us know already. Anyways goodjob to everyone helping out on both projects recently, just know a lot of people appreciate it!

maybe i messed up my test, but adptiveqos/flexqos was unable to control my latency when UL was saturated. no packet loss, but latency was over 1000ms