Hello! I hope you guys are doing well :>
I wanted to you guys an update that for me disabling these works for me and no longer crashes and even if does which is rare after 2 days or so
Currently everything is default with Guest Network made and Rate Limit from QoSmate to the whole Guest Network Subnet and that's it
QOSmate is unstable. Iâm out. Love the concept of hybrid mode, but it just crash on me on 24.10.6. Uptime ~ 7 days before it crashed. And no, it isnât hardware issue â before using qosmate, I have 100% uptime. Only downtime is upgrading the kernel when I saw something on the git logs that was useful.
Some notes: Loaded 24.10.6 as experiment on Thursday, (pass through 1 day soak test). Made it through easter weekend where wives and their munchkins doing their egg hunt. So lots of bursty traffic (social media) interleaved with regular downloads. Had high hopes since it held up. But it was for nought.
Why HFSC only lolcow? Go to linkedin, ask the head tech guy of FAANG (corporation=stand in for scaling), âhey man, do you use HFSC only on your networks for your corporation?â Or ask libreqos head tech guy, âhey man, why donât you use HFSC only instead of eBPF for your customer solutionsâ. Why did people invent other tools if HFSC exists? If youâre still using HFSC only for your network, youâre NGMI in 2026. Hybrid mode qosmate - ok. HFSC in 2008 â ok.
@AyanZ good luck with QOSmate! I just saw your disabling solutions, maybe it will lead to some solution thatâll make it 100% stable.
The way in which enterprise or ISP level platforms handle their queues has basically nothing to do with anything involving OpenWrt. In a Netflix datacenter at 200Gbps you could have a 1000 packet pfifo and it'd get drained in 0.06 ms. Noone needs queue management above 10Gbps really because you just forward packets as fast as possible and maybe if you accidentally build up a 1000 packet queue it causes an extra 1.2ms of delay. You can get super sloppy at speeds where an extra 1000 packets here or there is totally un-noticeable.
Where queue management really matters is for the people at 1Gbps down to like 1Mbps, and the slower your connection the more precise your queue management needs to be. A single extra packet getting sent too early at 1Mbps is an extra 12ms of delay.
Cake is designed to be "a piece of cake" to set up, and to reduce latency on everything equally to an acceptable target level of around 5 or 10ms. HFSC is designed to give you specific control over what kind of latency tradeoffs you will accept. For example, in order to keep your realtime control traffic down in the 0.5ms range while a large burst (say a world update) comes through you may accept 20 or 30 ms of delay increase on torrent traffic or long running game downloads.
Now, if there are bugs in the kernel code which results in "crash" then that's a kernel bug, which may be specific to certain hardware or to certain combinations, such as hybrid mode. I can tell you that for over a decade I've used HFSC in combo with pfifo and fq_codel and sometimes qfq and have never once had a crash associated with it. So if there are common crashes, and it involves hybrid mode, then I am guessing there's an interaction at the kernel level between HFSC and Cake, and this should get debugged but sounds like it's hard to reproduce.
In any case, do what you want, but for the other users, there's no need to be concerned that somehow QoSmate has out of date technology. The FFT algorithm was invented in 1800's and popularized in 1965 and it's still the basis for essentially all image and audio compression today.
Odd argument... LibreQoS uses HTB+eBPF+cake/fq-codel, but their usecase does not care for much in the traffic shaper part (they want need hard per subscriber limits and borrowing to allow for the oversubscription that ISPs use for several 100 users), so HTB is just as good as HFSC but HTB seems better maintained, if only because google seems to be using and maintaining it. Now, one can make the argument that that difference in maintainer activity is an argument against using HFSC, or one can accept that using HFSC might mean the occasional kernel bug hunting.
BTW, in Linux CBQ, TBF and HTB all preceed HFSC, so a lot of legacy users had already committed before HFSC arrived...
@dlakelan I apologize. I offended your izzat (you remind me of Indian posters).
Since HFSC is your âbaby so to speakâ relative to QOS, jogging my memory â some tik rates or whatever (I donât game). Now my memory is kinda running, I kinda got baited because I forgot â you are the OG IPV6 acoloyte and head priest spreading the goodness of IPV6, Chinese infiltration specialist (some boomer tier slop posting even here on openwrt, like the âthe registerâ LOL), even Iranian dissendent expert, and now youâre writing essays and so forth.
TL;DR: I agree use what you want
Right then I think that clears it up pretty well.
This is like wordcelling to the 10th degree. So they don't use HSFC only? Thanks for letting me know!
But I shouldn't be so sarcastic, you are after all, defending a friendship since you perceive I loudly challenge and in your mind, tastelessly...
Vibe on this website has shifted. Time to move on.
Who does not use HFSC only?
BTW had to look up wordcelling, not clear you are using the term correctly though... But let me ask, these hypothetical questions for linkedin, did you actually ask them, or are these pure speculations?
Sorry, what? I am trying to answer what I assumed to be bona fide questions. Apparently these where just rhetorical?
If you must, you must...
It turns out people who don't have a need for HFSC also don't use it, people like Netflix's core routers operating at hundreds of gigabits apparently. And since they don't use it, obviously we shouldn't either. Because Linkedin! /sarcasm
It's sad how much of a cult the world is these days. I think of OpenWrt as a part of the resistance though. We just keep making things that work well, and giving them away for free to people who will listen. James C Scott would be proud.
Does anyone know how I can extract what ports are being used for my game when playing it? Iâm trying to find what udp ports are being used for gameplay when in a match. Wondering if thatâs possible with geomate or Qosmate?
A packet capture will tell you this. however you should play several games to see if it varies from session to session.
