CakeQOS CakeQoS-Merlin v2.1.1

[privacyguy123]

No with this option you stick with the default (wan upload and ifb created for download based on wan). You just need to end up with 'flows nonat' by either tweaking the options and/or specifying custom upload parameters. And you can verify all worked well using 'tc qdisc ls' when logged in via ssh.

Cake automatically creates "ifb" ?

 

[dave14305]

Does the priority queue dropdown include the 'flows' option?

Yes, it does.

Cake automatically creates "ifb" ?

Yes, it does.

 

[privacyguy123]

Thanks, appreciate the noob explanations

Here's what I end up with:

tc qdisc ls
qdisc pfifo_fast 0: dev eth0 root refcnt 2 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1
qdisc pfifo_fast 0: dev eth1 root refcnt 2 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1
qdisc pfifo_fast 0: dev eth2 root refcnt 2 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1
qdisc pfifo_fast 0: dev eth3 root refcnt 2 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1
qdisc cake 8007: dev eth4 root refcnt 2 bandwidth 18432Kbit diffserv3 flows nonat nowash ack-filter split-gso rtt 100ms noatm overhead 80
qdisc ingress ffff: dev eth4 parent ffff:fff1 ----------------
qdisc pfifo_fast 0: dev eth5 root refcnt 2 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1
qdisc pfifo_fast 0: dev eth6 root refcnt 2 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1
qdisc cake 8008: dev ifb4eth4 root refcnt 2 bandwidth 110592Kbit besteffort dual-dsthost nat wash ingress no-ack-filter split-gso rtt 100ms noatm overhead 80

Is MPU important? It's set to 0 currently for me.

 

[Derek Zeitler]

Trying to remove Cake-QOS Merlin script with cake-qos uninstall but the tab is still showing under qos and Putty shows cake qos not found.

I am now trying Flex Qos and installed this instead.

I also cleared my cache and cookies still there.

Thanks in advance for your help

 

[Kingp1n]

Trying to remove Cake-QOS Merlin script with cake-qos uninstall but the tab is still showing under qos and Putty shows cake qos not found.

I also cleared my cache and cookies still there.

Thanks in advance for your help

You can try the following and reboot router:

CakeQOS - Uninstall current Cake Version to install new version

Hello.. I installed a version of Cake into merlin (RT-AX88 : 386.5_2) manually (I was rushing and haven't properly read all the instructions..) .. now I realized that I don't have the latest Cake V2.x.x due the GUI that shows on newer threads so I'm trying to remove it but not successfully... I...

www.snbforums.com

or

cake-qos uninstall

 

[Derek Zeitler]

You can try the following and reboot router:

CakeQOS - Uninstall current Cake Version to install new version

Hello.. I installed a version of Cake into merlin (RT-AX88 : 386.5_2) manually (I was rushing and haven't properly read all the instructions..) .. now I realized that I don't have the latest Cake V2.x.x due the GUI that shows on newer threads so I'm trying to remove it but not successfully... I...

www.snbforums.com

or

cake-qos uninstall

That worked thanks

 

[Lynx]

@privacyguy123 I've just asked a question on your behalf directed to one of the original cake co-authors on the OpenWrt forum:

How to use OpenWrt, SQM, and VPN on multiple connected routers?

@moeller0 I am trying to help someone using Asus Merlin get cake to work properly in the context of a WireGuard setup - see here: Looks like he now has 'flows nonat' + diffserv3. Will the latter kill the desired flow differentiation despite encryption? What effect will diffserv3 have when...

forum.openwrt.org

Specifically I think you may need to get rid of the 'diffserv3' keyword, but I'm not entirely certain. Hence my question.

I you run two concurrent speed tests via the VPN from two different clients, do you see 50/50 split of bandwidth?

 

[privacyguy123]

@privacyguy123 I've just asked a question on your behalf directed to one of the original cake co-authors on the OpenWrt forum:

How to use OpenWrt, SQM, and VPN on multiple connected routers?

@moeller0 I am trying to help someone using Asus Merlin get cake to work properly in the context of a WireGuard setup - see here: Looks like he now has 'flows nonat' + diffserv3. Will the latter kill the desired flow differentiation despite encryption? What effect will diffserv3 have when...

forum.openwrt.org

Specifically I think you may need to get rid of the 'diffserv3' keyword, but I'm not entirely certain. Hence my question.

I you run two concurrent speed tests via the VPN from two different clients, do you see 50/50 split of bandwidth?

Thanks! Please note it is possible to change diffserv3 to besteffort, difserv4 or difserv8

No, I get one hogging bandwidth as far as I can see ... also ping timeouts on the one that does the test slower.

Keep in mind my set up is ALL traffic sent though WG via VPN Director.

 

[dave14305]

Specifically I think you may need to get rid of the 'diffserv3' keyword, but I'm not entirely certain. Hence my question.

Isn’t the trade-off that WG will retain the skb hash in order to differentiate flows, but the encapsulated packet presents no original DSCP, so the tin would end up as whichever DSCP is assigned to the tunnel connection itself? besteffort is probably the most sensible choice. You can have fairness but not priority in such a scenario.

 

[Morris]

Isn’t the trade-off that WG will retain the skb hash in order to differentiate flows, but the encapsulated packet presents no original DSCP, so the tin would end up as whichever DSCP is assigned to the tunnel connection itself? besteffort is probably the most sensible choice. You can have fairness but not priority in such a scenario.

Not only this, you can tag your packets with DSCP yet will any of the routers along the way use them for queuing? Probably not.

 

[privacyguy123]

I honestly can't determine at all if QoS is doing anything on this router. Perhaps AX3000 is too old and doesn't have these features in the kernel or something? Bufferbloat tests online pretty much suck across the board, FlexQoS, Cake QoS or stock. running speed tests causes running video stream to become choppy or cut out entirely for up to a second *sometimes*, it is hard to reproduce but happens often enough. Ping tests ms shoots up high and sometimes it times out when running a speed test - I've seen others online say these settings are supposed to fix that.

Ironically upload on these tests is A+.

 

[Morris]

I honestly can't determine at all if QoS is doing anything on this router. Perhaps AX3000 is too old and doesn't have these features in the kernel or something? Bufferbloat tests online pretty much suck across the board, FlexQoS, Cake QoS or stock. running speed tests causes running video stream to become choppy or cut out entirely for up to a second *sometimes*, it is hard to reproduce but happens often enough. Ping tests ms shoots up high and sometimes it times out when running a speed test - I've seen others online say these settings are supposed to fix that.

Ironically upload on these tests is A+.

What is your measured download speed with no activity or as little as possible? Set 90% of that for CAKE and you should see a marked improvement. Also, on the CakeQOS-Merlin form, change your upload to match the download. When you move away from besteffort you are not following the spirit of the CAKE design and the same goes for diffserve3. For both download and upload you want the following:
NAT Lookup: Yes
Wash DSCP Markings: No (You don't want CAKE to use them yet upstream routers can)
Filter Duplicate TCP ACKs: No (Let TCP handle this tiny bit of traffic that is there if there is a problem)

Good luck

 

[privacyguy123]

What is your measured download speed with no activity or as little as possible? Set 90% of that for CAKE and you should see a marked improvement. Also, on the CakeQOS-Merlin form, change your upload to match the download. When you move away from besteffort you are not following the spirit of the CAKE design and the same goes for diffserve3. For both download and upload you want the following:
NAT Lookup: Yes
Wash DSCP Markings: No (You don't want CAKE to use them yet upstream routers can)
Filter Duplicate TCP ACKs: No (Let TCP handle this tiny bit of traffic that is there if there is a problem)

Good luck

120mb/20mb - VPN on or off both the same.

I was told by others that for Cake to play nice with WireGuard it has to be flows nonat. None of this matters as QoS does nothing discernable whatsoever on my connection with everything routed through WGC1 via VPN Director. Flex or Cake.

 

[Morris]

120mb/20mb - VPN on or off both the same.

I was told by others that for Cake to play nice with WireGuard it has to be flows nonat. None of this matters as QoS does nothing discernable whatsoever on my connection with everything routed through WGC1 via VPN Director. Flex or Cake.

WireGuard is probably a single flow as that's how VPNs work. All of the WireGuard traffic will be treated the same. What applications are running over Wireguard? If you have applications that require priority you will need to separate them for QOS to help. This means not using a VPN for them, using a deferent VPN for them or using something like Tailescale which uses separate flows for each application.

 

[privacyguy123]

As a last hail-mary-I-don't-know-wtf-is-going on I limited both sides to 20mb - even capped download speed at 20mbs is showing over 60ms latency added when downloading. how the F does that make sense? That's like 20% of my bandwidth and it's still bloating ...

 

[Morris]

As a last hail-mary-I-don't-know-wtf-is-going on I limited both sides to 20mb - even capped download speed at 20mbs is showing over 60ms latency added when downloading. how the F does that make sense? That's like 20% of my bandwidth and it's still bloating ...

Your frustration is clear. A VPN adds latency. Separate latency sensitive applications from the VPN. If they require encryption, then encrypt at the application level. Using a VPN is like putting everyone in a car. Do you expect some passengers to arrive sooner than others? If you want some passengers to arrive sooner, then they must be in a separate car that has flashing lights to give it priority.

 

[privacyguy123]

Your frustration is clear. A VPN adds latency. Separate latency sensitive applications from the VPN. If they require encryption, then encrypt at the application level. Using a VPN is like putting everyone in a car. Do you expect some passengers to arrive sooner than others? If you want some passengers to arrive sooner, then they must be in a separate car that has flashing lights to give it priority.

I am to believe that CakeQoS set how @Lynx describes sort those people into "different cars" ...

 

[Morris]

I am to believe that CakeQoS set how @Lynx describes sort those people into "different cars" ...

No. You are routing all traffic into a VPN. That's one car.

 

[dave14305]

I am to believe that CakeQoS set how @Lynx describes sort those people into "different cars" ...

It turns out that the CAKE source used in these routers is too old to have the capability to recognize the individual flows from Wireguard.

[Cake] [PATCH net] sch_cake: Take advantage of skb->hash where appropriate

 

[privacyguy123]

It turns out that the CAKE source used in these routers is too old to have the capability to recognize the individual flows from Wireguard.

[Cake] [PATCH net] sch_cake: Take advantage of skb->hash where appropriate

Things starting to make sense now.

Can this feature ever be "hacked" into our routers or firmware?