Active Congestion Control additional configuration

[pbix]

In this case your variation is beyond what ACC is designed to do. Its interesting to hear a case such as yours because its an extreme example of why ACC is needed.

There are reasons for the limit, its not arbitrary. If you set your bandwidth to 2 / 0.15 = 13mbps you will not have problems with QOS. If you do not care if QoS sometimes does not work you can maintain your current settings. I know you would prefer to have it all but this is the current state of the art.

If you know another router firmware that can do better it would be an interesting conversation as to how they do it.

[DasMarx]

Hello Eric, hello Lantis,

i will try to compile it myself and issue a pull request if i succeed.

Re Eric:

Background information for my line:
I use Deutsche Telekom Hybrid access which does combine a fixed DSL speed (2mbit) + variable LTE (0 - 70 mbit) to a combined line with a server in a datacenter. There is no data cap but the LTE part is lower prioritized than other LTE users (e.g. mobile phones).

Now regarding your questions:
1)
Yes, i did them while downloading.
The first one with a ping beeing shown and current link speed around 20 mbit.
The second one with no ping beeing shown and a link speed around 2 mbit where i disabled the LTE part of my connection

2)
Here are some background information how this line works:
https://tools.ietf.org/html/draft-lhwxz ... tecture-02
http://www.ietf.org/proceedings/90/slid ... enet-2.pdf

3)
I did test them either together (combined) or DSL alone to see how it performs when manually beeing slowed down to not beeing aggregated.
It is somehow different than cable access as my ISP does not promise any of the LTE part. The issue would be resolved, if the DSL part would be near or equal to the LTE part, but that is just a wish for the future....

4)
I try to test ACC and QOS the same way ACC works. I do ping tests to at least 2 systems that perform well (google and a local big news site normally).

I choose ping tests as a measurement for this issue, as it is the same method, ACC is using to decide if the current maximum download speed should be decreased or not.

I have not set many QOS classes but i put my NAS to download files into the slowest class with only 1% and every other system into a faster class. The equal distribution of QOS normally works good for me even when downloads occure.

My first test would be to issue some big downloads on my NAS and observe the ping timing on my workstation and in the gargoyle router.

Even at high download speeds and when ACC is active, my ping normaly does not spike above 100 to 140 ms. The ping is normally my way to see if QOS works. Other than that, i also open some websites and look at the individual timings of the chrome console or observe how the images build up. When QOS and ACC work, there is nearly no delay even on a congested line.

I see ping peaks up to 600 and 800 while ACC kicks in until the line is at its true speed.

If i disable my LTE part, my line is automaticly restricted to 2 mbit.
When i do the same test as above and download some big files with my NAS, my ping will instantly go to 800 - 1500 ms.
ACC is not active and does not show a ping.
Opening websites does not realy work even though my connection is in class "fast" and the NAS is in class "slow".
The line feels very laggy and additional streams on other devices like internet radio do not work anymore.

5)
Of course all tests are done with ethernet


re pbix:
I see the idea behind the limits, but they always assume that a line normally does never go below 10 to 15% of its maximum.
Also it is true that i can either set my download speed to a lower value to keep QOS intact or keep the higher value to get maximum download speeds most times of the day and not working QOS in the evening.

This is what i currently do, i keep the values high and just push the downloads to the night or morning when no one is using it.

[Lantis]

Let us know how you go.
For future, pbix implemented the ACC system so he is definitely the authority on this stuff. His information should be regarded well above my own.

[ericwong]

Background information for my line:
I use Deutsche Telekom Hybrid access which does combine a fixed DSL speed (2mbit) + variable LTE (0 - 70 mbit) to a combined line with a server in a datacenter. There is no data cap but the LTE part is lower prioritized than other LTE users (e.g. mobile phones).

This is certainly the first time I learned there is such service provided by ISP.
I had seen some 3rd party company providing hybrid access to boost internet speed but not from the ISP directly.
From where I live, the focus is mostly to upgrade to fibre.

It certainly looks like you have done a thorough job testing ACC and the problem seems to be the inability of ACC to truly measure bandwidth. In my opinion, such problem won't exist if there is a way for ACC to measure bandwidth without any user input

If the code change attempt didn't work, one thing I can suggest is write a plugin that run speedtest at cron (user defined intervals). The plugin will simply adjust the ACC bandwidth setting automatically. This plugin should be useful to other users too because asking user to enter your connection bandwidth can be somewhat difficult. I know there are scripts/ways to run speedtest on Openwrt, so such plugin should be feasible without too much work. The only problem is some QoS rule would be needed to prioritize speedtest bandwidth over everything else when this script runs.

e.g. https://forum.openwrt.org/viewtopic.php?id=46168

just push the downloads to the night or morning when no one is using it.

This is certainly the best way though