[RT-AC86U] I built cake!!!

[Wade Coxon]

cake+suricata 250Mbit 66% CPU
cake only 300 Mbit 25%
no cake no suricata 300 Mbit 25%

all via wifi.
speed test from router is saying 750Mbit and more ...

It would be worth reporting what you get from a wired client connected at 1Gbps.

Good to know that 300Mbit seems to be more than achievable though!

 

[Milan]

It would be worth reporting what you get from a wired client connected at 1Gbps.

Good to know that 300Mbit seems to be more than achievable though!

first two are same , third is 600 Mbit on 1Gbit cable - testedon ubuntu iso download.

 

[ttgapers]

Interesting project! Sadly I think my 350/35 line is probably on the high side for Cake? I saw somewhere in the thread that over 250 wasn't recommended

I should mark that as a "guideline" more than a recommendation. I'd give it a whirl!

 

[ttgapers]

@ttgapers i wonder if its possible to tie in spdmerlin to pass speedtest results to cake script ....cron job? @Jack Yaz @jackiechun

TBH, I don't think Cake "needs" this per se, as I mentioned it "just works" as adding the speedtest with it, may impact the way it performs as it "believes" your upper limit might be lower than it actually is based on when spdmerlin ran vs. Cake load situation at any given point in time. I am not a Cake expert but didn't consider the AutoBW mod due to the same reason (I could potentially lower my line speed if it has a "bad" run).

Thoughts from the elders appreciated on this...

 

[ttgapers]

It would be worth reporting what you get from a wired client connected at 1Gbps.

Good to know that 300Mbit seems to be more than achievable though!

I added my screenshot from a wired client to my post on Page 5.

 

[ttgapers]

I will work on a new thread after work today. Let's see if it gets some more testing for the next 12-16 hours.

 

[TechTinkerer]

TBH, I don't think Cake "needs" this per se, as I mentioned it "just works" as adding the speedtest with it, may impact the way it performs as it "believes" your upper limit might be lower than it actually is based on when spdmerlin ran vs. Cake load situation at any given point in time. I am not a Cake expert but didn't consider the AutoBW mod due to the same reason (I could potentially lower my line speed if it has a "bad" run).

Thoughts from the elders appreciated on this...

Maybe a specific number of tests could be run then the average speeds with 95% calculated then passed to the cake script

 

[Wade Coxon]

first two are same , third is 600 Mbit on 1Gbit cable - testedon ubuntu iso download.

Interesting. So if the CPU figures were the same, that's a hard 300Mbps limit without needing out the CPU.

Was that CPU utilisation for a single core or combined? If have thought that when cake ran out of puff, it would have completely maxed out one core.

 

[Wade Coxon]

I added my screenshot from a wired client to my post on Page 5.

That's not on a gig WAN connection though is it? I'm just asking the question about a wired client on a gig WAN connection so we can eliminate them as the bottlenecks and measure the performance of cake on these Broadcom chipsets.

 

[ttgapers]

That's not on a gig WAN connection though is it? I'm just asking the question about a wired client on a gig WAN connection so we can eliminate them as the bottlenecks and measure the performance of cake on these Broadcom chipsets.

Correct on all points....

 

[Milan]

Interesting. So if the CPU figures were the same, that's a hard 300Mbps limit without needing out the CPU.

Was that CPU utilisation for a single core or combined? If have thought that when cake ran out of puff, it would have completely maxed out one core.

it was combined. no single core blow out, there was some load on all cores for Suricata case. for cake only case, i can't say if it was only one core as no process shown with some significant load in htop.

 

[randomName]

Can it be installed without Entware? If not, why is it needed?

 

[Wade Coxon]

Can it be installed without Entware? If not, why is it needed?

The kernel modules are packed in ipk files, so you need opkg to manage the installation of those.
In theory you could do it without them, but it'd be messy I suppose.

 

[jackiechun]

I will work on a new thread after work today. Let's see if it gets some more testing for the next 12-16 hours.

In the script I think 5m.ca/cake/ needs to be replaced with raw.githubusercontent.com/ttgapers/cakeqos-merlin/master/, and a delay might be needed to the beginning of the "start" section because it's not starting up properly when I reboot my AC86U (FreshJR has a delay of 5 minutes, sleep 300s). It's working fine if I manually stop and start.

 

[ttgapers]

In the script I think 5m.ca/cake/ needs to be replaced with raw.githubusercontent.com/ttgapers/cakeqos-merlin/master/, and a delay might be needed to the beginning of the "start" section because it's not starting up properly when I reboot my AC86U (FreshJR has a delay of 5 minutes, sleep 300s). It's working fine if I manually stop and start.

Do you have a github account? I can give you access if you make the request. Might also be a good time to clone a develop branch off of master. Let me know, but I am strapped for time until later tonight EST to make any edits....

Secondly, I do want to switch the ipk links to github, but had random issues last night (filesize etc. not correct), so in the interest of time I left it where you host it. That is another quick win for future updates....

 

[Wade Coxon]

In the script I think 5m.ca/cake/ needs to be replaced with raw.githubusercontent.com/ttgapers/cakeqos-merlin/master/, and a delay might be needed to the beginning of the "start" section because it's not starting up properly when I reboot my AC86U (FreshJR has a delay of 5 minutes, sleep 300s). It's working fine if I manually stop and start.

Should it be linked to WAN start rather than the device startup?

 

[ttgapers]

Should it be linked to WAN start rather than the device startup?

This is not a bad idea...thoughts from the others?

 

[jackiechun]

Do you have a github account? I can give you access if you make the request. Might also be a good time to clone a develop branch off of master. Let me know, but I am strapped for time until later tonight EST to make any edits....

Secondly, I do want to switch the ipk links to github, but had random issues last night (filesize etc. not correct), so in the interest of time I left it where you host it. That is another quick win for future updates....

Thanks, will do later tonight (busy work day!)

Should it be linked to WAN start rather than the device startup?

That's a good idea; I also just realized FreshJR's QoS is attached to firewall-start (+ 5 minute delay). I don't know the boot sequencing too well so hopefully others would be able to suggest where to attach this script! (services-start, wan-event, firewall-start, or another).

https://github.com/RMerl/asuswrt-merlin.ng/wiki/User-scripts#available-scripts

Sounds like wan-event might be a little dangerous because it seems like it'd be triggered if the connection drops/reconnects. Perhaps firewall-start is the way to go. "Called after the firewall got started and filtering rules have been applied. This is where you will usually put your own custom rules in the filter table (but not the NAT table)."

Has cake-qos been auto-starting properly for anyone, with the services-start entry?

 

[heysoundude]

This is not a bad idea...thoughts from the others?

I've not played with anything on the WAN-start side of things, but a 300 second nap seems excessive to me.
I suspect/feel that surely 60-90 seconds in most cases would be (more than?) sufficient if it were placed correctly in the startup sequence, but I defer to those who know better than I how that all happens...
come to think of it, not everyone does their ISP sign-in on their router, so that would definitely change timing/sequencing...and I'm using DDNS that has to wait for my PPPoE to authenticate with my ISP and get an IP addy; so in my case a router reboot takes possibly as much as 90-120 sec (I've never timed it).
I'm also not a script-savvy bloke, but maybe there's an if-then logic that can be applied to starting cake when certain conditions are met?

 

[ttgapers]

Sounds like firewall-start might be best....