search

LibreQoE / LibreQoS

A Quality of Experience and Smart Queue Management system for ISPs. Leverage CAKE to improve network responsiveness, enforce bandwidth plans, and reduce bufferbloat.
https://libreqos.io/
GNU General Public License v2.0
439 stars 47 forks source link

QosIfy - dynamically remarking DSCPs via ebpf #113

Open dtaht opened 2 years ago

dtaht commented 2 years ago

Over here there is a clean module for recognizing identifiable traffic and applying diffserv markings to it, for cake. https://forum.openwrt.org/t/qosify-new-package-for-dscp-marking-cake/111789

I am no fan of dscp, and of the near religious fights over what each codepoint means. Also, many tests to date of linux's wifi code show how badly those interpretations "fit" into wifi's access classes (which govern access to the media, not flow behavior). As best as I can tell many older, and bigger ISPs use (or used to use it) to remark traffic and differentiate between voip, business, and residential traffic on a strict priority queue basis, and didn't (among other things) put CS1 into a background queue properly (which is why I actually am a proponent of the new LE codepoint in https://datatracker.ietf.org/doc/html/rfc8622 )

I as a user, would like very much to toss backups, and torrent, for example, into the background.

There are other, saner means of doing traffic engineering, and if I had any hope for DSCPs was that they would actually be preserved e2e one day and more often. So far I primarily am observing that dscps are remarkably nowadays preserved e2e, and notably at least some cell phones are setting the right ones to land in the right tins in cake.

So, anyway cake does soft admission control for the classes it supports, rather than strict priority and is thus less dangerous to use.

So IMHO - on the whole FQ is, like 95% of a good solution, AQM 3%, ECN potentially another 1%, and the rest... magic. But I could be wrong, particularly when it comes to videoconferencing. I think strict priority queues exist, still, in a lot a gear, and so long as you are doing soft admission control a la cake, some marking might make sense, and taking stock of any improvements to how zoom might work, in particular has long been on my mind.

https://support.zoom.us/hc/en-us/articles/207368756-Using-QoS-DSCP-Marking

dtaht commented 1 year ago

I would like us to better observe actual DSCP mark usage in the wild, and whether or not they help.