FreeBSD 14 supports some of the more important parts of L4S, such as AccECN. It is missing "TCP Prague" but having L4S support for UDP is really the bigger use-case.
I wasn't able to find out if FreeBSD has added any of the new AQM implementations L4s refers to, such as DualPI2, so maybe we have to wait on that part to land in FreeBSD first?
Anyway, based on the white paper listed below L4S provides much better queueing delays than the AQM implementations available in OPNSense today and interested in trying it out soon.
References
FreeBSD 14 supports parts of L4s:
https://freebsdfoundation.org/our-work/journal/browser-based-edition/networking-10th-anniversary/updates-on-tcp-in-freebsd-14/
L4S White Paper:
https://www.bell-labs.com/institute/white-papers/l4s-low-latency-low-loss-and-scalable-throughput/
For starters, you will have to wait until most backbone providers support that. On page 12 of the underlying paper, this is clearly stated:
Quote
L4S requires two parts to work: 1) senders and receivers with L4S-capable congestion control in the
application; and 2) L4S AQMs and isolation mechanisms deployed in the bottleneck node(s) on the
end-to-end path within the network. It is not sufficient to have only one of these two parts. Application
providers and network operators, therefore, each hold one half of the key to enabling the performance
benefits of L4S.
Part of this is that ECN bits be transferred from end to end.
So, while it is nice to see that there will be progress on this by incorporating it into widely deployed platforms like FreeBSD, there is yet little benefit to support it in OpnSense - apart from lab setups, that is, but these do neither suffer from high latency nor packet loss or congestion unless synthetically induced.
Apart from that, OpnSense usually is neither a client nor a server, only a router - apart from cases where it hosts a forward or reverse proxy. As a router it just forwards packets and has little impact on timing (modulo traffic shaping).
^
|
Couldn't agree more. Also lets not forget here for ISP to implement something like that takes ages. A lot of the ISPs didn't yet even implement LibreQos.... Or any AQM/SQM for that matter, just running FIFO with BW limitations.
OPNsense e.g FreeBSD as such would more benefit if it had a full implementation of CAKE, so we can take advantage of at least to try to control the buffer bloat from our side.
Regards,
S.
QuotePart of this is that ECN bits be transferred from end to end.
ECN bleaching is a problem, but seems this is pretty low, 88% keep the flag over internet traffic, according to Catchpoint's tests (https://www.catchpoint.com/reports/is-the-internet-ready-for-l4s).
QuoteAlso lets not forget here for ISP to implement something like that takes ages. A lot of the ISPs didn't yet even implement LibreQos.... Or any AQM/SQM for that matter, just running FIFO with BW limitations.
Ya the bandwidth limit enforced by the ISP is almost always going to be the default bottleneck, since we are always trying to get the most out of the bandwidth we are paying for. Since we can't control the ISP's queueing logic what I believe is done with basically all the AQM/SQM strategies at the router level is to go under the bandwidth provided by your ISP so it's more effective. So while it would be helpful if L4S was also implemented by the ISP I don't think it is needed to get most of the benefits of L4S; as long it the app, its server, and the main bottleneck all support L4S.
The main problems with ISPs (at least in EU), is the fact they just use plain FIFO. FIFO is not cutting it, never was, and should be eradicated. There are several options how ISPs could improve their network in regards of latency e.g buffer-bloat and would not cost them anything or not much at least. The funny bit about this is that a lot of ISPs dont even know the term bufferbloat/SQM/AQM (and I am not speaking about the small ones).
For FreeBSD and OPNsense I hope in regards of this topic for 2 things:
1. To have the annoying FQ_CoDel BUG/implementation FIXED in FreeBSD
2. To have CAKE implemented into FreeBSD and OPNsense taking advantage of it.
Regards,
S.