Recommend a new router for Gargoyle?

[fifonik]

I tried Gargoyle 1.10..

There is no such version.
There are 1.10.0 (last stable), 1.10.x (was in dev after release of 1.10.0 then becomes 1.11.x) or 1.11.x (current dev)

I didn't think to turn off bandwidth monitor.

It was discussed in the thread that if you cannot reach speed required, switching off BM mon would be the first option to try as this feature is resource expensive.

I've upgraded to 400/20

turning on Gargoyle's QoS with the default settings and the upload bandwidth set to 450000kbps then 400000 then 350000.

Did you mean download bandwidth?
It looks like you are just playing with numbers.
There is no any reason to specify QoS BW as 450 when your connection is only 400.

[gelliss]

I believe I said the last stable version.

I just forgot to try turning off bandwidth monitor since you have to log in with putty to do that. As one of the moderators said in another thread when I believe you or someone suggested adding a setting on the GUI to turn it off, that kind of defeats one of the main features of Gargoyle. I may try it again when I have more time and will turn it off. I did that on my old C7 router but didn't see much difference.

And yes, I meant download bandwidth. I already said Spectrum over-provisions the cable modems to 480/23mbps rather than the advertised 400/20. I tried setting download bandwidth from 450000 down to 350000kbps but using Gargoyle's default QoS settings, nothing changed. It was still well under 200000kbps down on speedtests with QoS enabled. Without QoS, Gargoyle does the full 480000/23000 kbps with no problem.

I just thought the developers might like knowing what I was seeing with this particular router. Lantis already said Gargoyl's QoS uses an entirely different algorithm. If not, that's cool. I'll stop posting about it.

[Lantis]

I take all of these reports seriously.
And I do genuinely intend to test most of them.

However, I have 8 routers on my desk and all of them are in various stages of firmware development etc which makes it hard to test some things (like this). So I couldn’t give a timeframe, but I do genuinely take notice of all reports such as this.
Gargoyle can’t improve and move forward without the community, so I appreciate all feedback.

[fifonik]

gelliss@ I just wanted to clarify as some people asking about HW requirements for fast connections and in the forum we only have rumors/guessings.

As you have fast connection & powerful router it would be nice if you install the latest gargoyle (1.11.x RC1 and RC2 are available) and report speeds with BW ON + QoS OFF (default, as you already reported for 1.10.0), BW ON + QoS ON (as you already reported for 1.10.0), BW OFF + QoS ON.
You should not see any speed changes with BW OFF + QoS OFF as it is utilizing your connection with QoS only is OFF.
Thanks in advance.

[gelliss]

I reinstalled Gargoyle 1.11.x from here: http://lantisproject.com/gargoyle_custom/mvebu-1.11.x/
I was able to go from openwrt 18.06.1 directly to Gargoyle using the system upgrade file there. I didn't try to save any settings.

I don't think that's the latest build but it's what I used for the testing purposes. I didn't have a link and it's not that easy finding these beta files from google.

I immediately got full speed on my WRTacs v.1. I tested three times on dslreports speed tests. Download ranged from 462 to 482 down and a high of 21.94 up. The results didn't vary much in three tests. Bufferbloat was D

With QoS turned on and using the defaults. I set download bandwidth to 400000 and upload to 20000. I rebooted the router.

I ran DSLspeed tests another 3 times and got between 188-198 down and a high of 19.27 up. Bufferbloat improves to A or A+

I then turned off bandwidth monitors using putty and the stop/disable commands:
/etc/init.d/bwmon_gargoyle stop
/etc/init.d/bwmon_gargoyle disable

Again rebooted the router.

DSL speed tests showed little gain from turning off BW monitors.

I ran another three tests and the results were
198-204 down and a high of 19.x up

A last test was with BW monitor still off but ACC checked in the QoS settings.

Speed dropped another 10mbps or more with 3 tests between 181-190 down. Upload still around 19.

I much prefer Gargoyle's interface and ease of use but will probably put Openwrt back if I'm concerned enough with bufferbloat that I want to run QoS. There is just too much of a difference in QoS performance between Openwrt and Gargoyle.

[Lantis]

I've completed some speed testing on a fresh wrt1900ac v1 and have confirmed similar numbers.

For the benefit of all readers, the raw processing speed (bwmon on, qos off) of this device is close to 800Mbps.

I'm now doing some research and testing on improving the other performance numbers, although this is not my area of expertise. I have reached out to some more knowledgeable members of the community for this.

[gelliss]

@lantis

Is it possible to give a brief (for us laymen) explanation of why there is such a big difference in QoS between Openwrt/Lede and Gargoyle? You just said it's a completely different algorithm. I'm just curious how it works but if it's too technical it would go away over my head. lol

Also, can you post a link to the latest builds of Gargoyle? The links to builds in your signature appear much older than the one I found. Fifonik mentions RC1 and RC2 builds but I don't know which they are.

[ektus]

On the forum main page (index.php) scroll down to the "Development" section (viewforum.php?f=10) and enter the "Show / Tell / Contribute" area (viewforum.php?f=14) , where you'll find threads on all the latest builds, usually with links in the first post of a thread.

[gelliss]

thank you, ektus

[Lantis]

I'm not as familiar with other implementations, and admittedly, not as familiar with the Gargoyle one as i should be.

SQM is the most popular algorithm in Openwrt which is more than just QoS. It employs several techniques like queue management, traffic shaping/policing, rate limiting and prioritisation. This is a very comprehensive solution and hence why it is so popular, and it works!

Gargoyle does the same but utilising some different queuing methods and disciplines. For example we use a method called SFQ (Stochastic Fairness Queuing). If you imagine 3 lines at a single ticket booth, and the operator calls someone from line A, then someone from Line B, then C, and back to A etc etc. Essentially, each line gets an equal turn, and hence why Gargoyle's QoS operates well under saturated links. It guarantees that all streams get a turn.
** take this next bit with a pinch of salt as i'll probably ruin this explanation **
My understanding is if you want to prioritise some traffic, or put a hard limit on some, you jut make their line shorter.
They still get called up to the ticket booth equally like everyone else, but if the line is full, people get turned away. Because people are turned away, the rate at which data is sent to that queue has to slow down or you'll just continue to disappoint them.


Now, why is Gargoyle's QoS having such an impact on performance? My theory is that the maximum lengths of the queues is not enough. There's some maths behind it which i haven't understood yet (and i've reached out to someone who knows, or at least used to), but the theoretical maximum through a single queue is essentially 350Mbps.
It might be as simple as lengthening the queue, or we might need to introduce additional queues, or maybe SFQ is not appropriate above a certain bandwidth limit. I'm not sure yet, but i'm looking.