Well of course, I can easily post 1ms times from a box in AWS London, you're never going to get that living elsewhere in the UK unless on a private circuit. I peer directly to London and I see ~5ms pings to London based hosts.
Virgin Media Discussion Thread
- Thread starter macca40
- Start date
@Rainmaker : nice to meet you as well. Yes, it didn't look like you were making an apples to apples comparison, thank you for clearing that up. Also, I am saddened that the monitoring site you use takes the average of 100 pings. There is no real information in the average, but as one delay out of a few dozen packets can make you miss your shot. So we tend to use cdf plots of *all* the data, and I do wish we could train the general populace to grok those.
I'm busy this weekend, some examples of use of our main tool (flent.org) are over here: https://forum.openwrt.org/t/validating-nss-fq-codels-correctness/111123/
The big question I'd had was wondering if VM had rolled out DOCSIS-pie yet, which is part of the DOCSIS 3.1 std. During COVID, Comcast did a *full rollout* across their entire userbase, PLEASE, EVERYONE IN THE UK: skip to the lovely graphs after pp 14. https://arxiv.org/pdf/2107.13968.pdf and beg VM to implement this if they haven't.
(the easy way to detect that would be to generate an artificial load for 5+ minutes (seeding a few popular torrents perhaps) and monitor with that broadband tool, or use flent to generate one.
As big a fan as I am of fq_codel and cake, and of taking control of your bandwidth away from your ISP...
if somehow, someday, your ISP managed bufferbloat right - especially also accounting for local congestion as various folk on this forum are complaining of, cake or pie could work better than any fixed parameterization you might hand your own router. Although we designed cake to be able to respond to signals from upstream, we don't get those from any ISP (as yet), and otherwise dynamically sensing changes in the bandwidth is *hard*, but see: https://forum.openwrt.org/t/cake-w-adaptive-bandwidth/108848/186 for some progress being made there, however.
Having given y'all some sunday evening reading, I have to note that I'm not the inventor of fq_codel or cake, they are my conception, not invention. I'm merely the one in the group that gives the most entertaining talk, and I set aside one day a week to google for fq_codel and post on some site I haven't posted good info on yet.
Full inventive credit belongs to kathie nichols and van jacobson for codel (rfc8289), eric dumazet for fq_codel(rfc8290), a team of 8 led by jon morton for cake, and toke hoiland jorgenson and michel kazior for fq_codel for wifi ( https://lwn.net/Articles/705884/ ) which is actually, and silently, one of the most widely adopted techniques in many APs today.
If anyone out there would like an entertaining introduction to how tcp actually works, and what bufferbloat is see: https://blog.apnic.net/2020/01/22/bufferbloat-may-be-solved-but-its-not-over-yet/
I gotta go off now and bang the drum for funding for an open source version of apple's new networkQuality test. https://github.com/network-quality/.../master/draft-cpaasch-ippm-responsiveness.txt
I'm busy this weekend, some examples of use of our main tool (flent.org) are over here: https://forum.openwrt.org/t/validating-nss-fq-codels-correctness/111123/
The big question I'd had was wondering if VM had rolled out DOCSIS-pie yet, which is part of the DOCSIS 3.1 std. During COVID, Comcast did a *full rollout* across their entire userbase, PLEASE, EVERYONE IN THE UK: skip to the lovely graphs after pp 14. https://arxiv.org/pdf/2107.13968.pdf and beg VM to implement this if they haven't.
(the easy way to detect that would be to generate an artificial load for 5+ minutes (seeding a few popular torrents perhaps) and monitor with that broadband tool, or use flent to generate one.
As big a fan as I am of fq_codel and cake, and of taking control of your bandwidth away from your ISP...
if somehow, someday, your ISP managed bufferbloat right - especially also accounting for local congestion as various folk on this forum are complaining of, cake or pie could work better than any fixed parameterization you might hand your own router. Although we designed cake to be able to respond to signals from upstream, we don't get those from any ISP (as yet), and otherwise dynamically sensing changes in the bandwidth is *hard*, but see: https://forum.openwrt.org/t/cake-w-adaptive-bandwidth/108848/186 for some progress being made there, however.
Having given y'all some sunday evening reading, I have to note that I'm not the inventor of fq_codel or cake, they are my conception, not invention. I'm merely the one in the group that gives the most entertaining talk, and I set aside one day a week to google for fq_codel and post on some site I haven't posted good info on yet.
Full inventive credit belongs to kathie nichols and van jacobson for codel (rfc8289), eric dumazet for fq_codel(rfc8290), a team of 8 led by jon morton for cake, and toke hoiland jorgenson and michel kazior for fq_codel for wifi ( https://lwn.net/Articles/705884/ ) which is actually, and silently, one of the most widely adopted techniques in many APs today.
If anyone out there would like an entertaining introduction to how tcp actually works, and what bufferbloat is see: https://blog.apnic.net/2020/01/22/bufferbloat-may-be-solved-but-its-not-over-yet/
I gotta go off now and bang the drum for funding for an open source version of apple's new networkQuality test. https://github.com/network-quality/.../master/draft-cpaasch-ippm-responsiveness.txt
Since I'm here today... what do you get under load? 1ms *idle* is meaningless, it's under normal working conditions (load) when the latency matters. This is what cake does, under load, on sonic fiber in san francisco, before and after. it had 80ms latency spikes, reduced to sub 2ms flat with cake.
http://www.taht.net/~d/sonic_106_cake_vs_default.png
Usually 2-3 ms extra but I'm not with VM, I've got FTTP with Zen.
I agree, and also use other tools when I need to actually troubleshoot something. Unfortunately, Virgin Media only tend to take notice of those ThinkBroadband BQMs. I've tried discussing iperf3, flent and netperf results before and it's just met with a blank stare, followed by 'Let's see a BQM'...
Way ahead of you there, Dave. VM aren't renowned for their listening skills, as I'm sure you're aware. Add to the list IPv6, TLS or at least not plaintext password entry on their 'hubs', and a myriad other things. Liberty Global seems to be better in that regard, so there's at least some hope. I'm hoping to move to FTTP in the next couple of years anyway, where I won't have to put up with as much nonsense (in theory). At present, VM's gigabit service has a single DOCSIS 3.1 downstream channel only. The rest (down and up) are DOCSIS 3.
Am I correct assuming that cake is still Linux-only? I moved away from using Linux (Debian, mostly) for my edge router a few years ago, and have been using OpenBSD. Those graphs are very persuasive... I may have to re-evaluate my choices!
Say the boogeyman's name three times, and he appears.
Very happy that you did. With the volume of interesting material to get my reading chops around, however, I'm not sure the same will be said of my wife!Thank you so much again for taking time to contribute and pass on those insights. There's a few of us here who have an interest in ('proper') networking equipment, protocols and operating systems, so I'm sure I'm not the only one who will benefit. Feel free to stick around (though I suspect not). As a brief aside, have I seen you on Computerphile's computer science videos at some point? I can't shake the feeling but may have mixed up my sources. Regardless, thanks again and have a good weekend - what's left.
I agree, and also use other tools when I need to actually troubleshoot something. Unfortunately, Virgin Media only tend to take notice of those ThinkBroadband BQMs. I've tried discussing iperf3, flent and netperf results before and it's just met with a blank stare, followed by 'Let's see a BQM'...
Way ahead of you there, Dave. VM aren't renowned for their listening skills, as I'm sure you're aware. Add to the list IPv6, TLS or at least not plaintext password entry on their 'hubs', and a myriad other things. Liberty Global seems to be better in that regard, so there's at least some hope. I'm hoping to move to FTTP in the next couple of years anyway, where I won't have to put up with as much nonsense (in theory). At present, VM's gigabit service has a single DOCSIS 3.1 downstream channel only. The rest (down and up) are DOCSIS 3.
Am I correct assuming that cake is still Linux-only? I moved away from using Linux (Debian, mostly) for my edge router a few years ago, and have been using OpenBSD. Those graphs are very persuasive... I may have to re-evaluate my choices!
Say the boogeyman's name three times, and he appears.
Thank you so much again for taking time to contribute and pass on those insights. There's a few of us here who have an interest in ('proper') networking equipment, protocols and operating systems, so I'm sure I'm not the only one who will benefit. Feel free to stick around (though I suspect not). As a brief aside, have I seen you on Computerphile's computer science videos at some point? I can't shake the feeling but may have mixed up my sources. Regardless, thanks again and have a good weekend - what's left.
I'm not good at web things. Responses in LIFO order:
I like to teach people to fish, rather than give them fish. So few know anything about queue theory, or even why it (and operations research) are so valuable, so lacking finding any jobs in my field on dice, I do this.
I have not been on computerphile. Should I? I'd really like to do some animations instead of talks.
In terms of sticking around, I try only to do the web thing on weekends. I like meeting new people tho, thx for being here, and every person that manages to fix bufferbloat makes for a better internet, especially if they then tell two more people how. Go forth and fix it for your mom, your local cafe, your school!
Most of the work on bufferbloat takes place on the mailing lists at https://lists.bufferbloat.net - I wish google indexed those, and more people were subscribed.
Perhaps you and your wife might enjoy my apnic talk together? In all of my talks I try to explain things while being entertaining and without a single equation.
Cake is still linux-only. Although we dual bsd-gpl'd it so as to ease a port to other OSes, I don't know of anyone working on it. There is a ns3 version nearing completion. We also designed cake to be implementable in hw, things like the 8 way set associative hash are std IP block, hashing and timestamping also, and there's been some good work on non gpl'd flow dissectors as well of late. Ideally the BSD's would gain an implementation of "BQL" as well.
I just looked at openbsd's fq_codel implementation. I'm a little concerned at a brief look at it as it appears to cut off codel's invsqrt calculation at count 400, where it should go to the max a u16 or u32 goes. I didn't look very hard but if you know of anyone over there...
People seem to forget openwrt runs on x86 also. It's the core R&D platform for the internet. Latest release is pretty good. Cake also went into linux mainline as of release 4.19.
Over here I explained how to use cake on the new mikrotik release: https://forum.mikrotik.com/viewtopic.php?t=179307
As for annoying VM enough? I just had a *ball* schooling starlink about bufferbloat. I told that (increasingly hilarious) 7 minute story about trying to file THAT bug report - having to call in vint to get to elon - over here:
https://www.youtube.com/watch?v=AjZXx4N1tmY&t=2680s
I was really befuddled that starlink shipped a 10 year old version of openwrt - and I need to be clear, cake belonged in the dishy, fq_codel for wifi in the router, but the starlink router was the only prop I had handy while losing my temper.
After that little talk... For all I know I am responsible for the some of the delays in launching their next sats and new dishy. I hope so. I'm going to get around to publishing the formal study at some point soon, some early data's over here: https://docs.google.com/document/d/1puRjUVxJ6cCv-rgQ_zn-jWZU9ae0jZbFATLf4PQKblM/edit -
Soooo.... Who do I nag about improving BQM? Instead of just an average they could display a CDF.
As for beating up Branson to fix bufferbloat... hmm... how could I do that?
Last edited:
What happened to Dave? I got a PM that I'm not sure was an insult or a compliment (though I'm leaning toward the former lol) and he's perma'd. Did he nuke his own account? You can't remove your own posts though, can you?
Edit: Having spoken to Dave on PM (that's still available to him), it's actually a funny situation. He's sent me a complimentary message but because he used a five letter US English term I can't repeat here (begins with s ends in k, means verve/courage/grit... but in the UK means man juice) the auto-mod starred it out. What was left sounded like an insult if you added other potential banned words in its place. He sent me a nice message, and the auto mod made it mean. The irony haha.
I'm RTM'ing this post. Admin/Dons, it seems Dave was banned but he has no idea why. He posted a reply with a lot of links (research, videos etc) and I think that, with his account being new, he's triggered the anti-spam-bot-bot? Can someone please check and reinstate his account provided that's the case? He's a world authority on networking, and we'd be silly to ban him by accident if that's what's happened...
Edit: Having spoken to Dave on PM (that's still available to him), it's actually a funny situation. He's sent me a complimentary message but because he used a five letter US English term I can't repeat here (begins with s ends in k, means verve/courage/grit... but in the UK means man juice) the auto-mod starred it out. What was left sounded like an insult if you added other potential banned words in its place. He sent me a nice message, and the auto mod made it mean. The irony haha.
I'm RTM'ing this post. Admin/Dons, it seems Dave was banned but he has no idea why. He posted a reply with a lot of links (research, videos etc) and I think that, with his account being new, he's triggered the anti-spam-bot-bot? Can someone please check and reinstate his account provided that's the case? He's a world authority on networking, and we'd be silly to ban him by accident if that's what's happened...
Last edited:
Associate
2028 will still be much quicker than Openreach will get to me I suspect.
Still hoping to be able to order 1Gig from Virgin before the end of the year
The last regions are set to be upgraded by the end of the year and they seem to be more of less on target to deliver that, traditionally they avoid doing anything significant like that just before/after Christmas/New Year, so realistically it should be pretty imminent.
Good morning from Amerika! Thx for restoring my access. I had tried to applaud @Rainmaker for his... spark... in going after VM on the DOCSIS-PIE issue - In my life I have been very inspired by British humor and ... pluck. (please note, I'm now googling these words as I go to see if they too are banned).
Monty Python's Life of Brian got me out of church, and the radio play of Hitchhiker's Guide to the Galaxy I've memorized, as we've (the bufferbloat fighting folk) have been trying to shift the global conversation about the internet for a decade now away from "Bandwidth" - towards the real factor that actually matters - reliably low round trip times, especially when the link is actually carrying traffic.
The billions being burned o fiber buildouts is one example - sure, you get more bandwidth but another core and underappreciated fiber feature is the vastly reduced - potential - round trip times. But unless you ALSO get better routers, the jitter and latency under working conditions can actually get worse!
Hitchhikers was BRILLIANT: "Why are you building a bypass?" - "Because bypasses MUST be built!" - and "The plans have been on file at alpha centuri for 50 years, I don't understand what you humans are so bothered about"... boom! earth destroyed. For no good reason. Similarly, when we'd made the breakthrough in queue theory that fq_codel was, fixing a problem that the internet had had since 1983, and anti-patented it and showed that few people actually needed more that 20Mbits of bandwidth so long as there was good queue management... and released the source code for free, we'd mistakenly thought the world would just do a little zero cost software upgrade and everybody, even and especially DSL folk, have better internet, all the time!
Perhaps this is the year some world leader or businessman would use bufferbloat coherently in a sentence.
On my guitar is a sticker that says THIS MACHINE KILLS VOGONS - which is a reference to the guide, and the American folksinger who wrote "This land is your land" - a song tied up in a copyright dispute.
Anyway, there's a ton of useful links in more formerly banned post above and I hope y'all can have a laugh at musk's expense. I figured they only had to fix 3 routers, by a recompile...
I was also in my youth very inspired by the honesty and directness of British punk, notably the Stranglers and Killing Joke. One of my best friends, based in bristol, Paul Whitrow died last year, not from covid, but apoplexy. I keep hoping for more Arthur Dents to ask difficult questions of regulators, and more folk "Get in there" to fix their internets for themselves, as we are all stuck, still, working from home.
A huge irony of the need for better videoconferencing is that videoconferencing is actually low bandwidth, it just needs reliably low jitter and queuing delay. Why can't the ISPs or regulators talk about that?
Because bandwidth must be built!
#FreeDave
Monty Python's Life of Brian got me out of church, and the radio play of Hitchhiker's Guide to the Galaxy I've memorized, as we've (the bufferbloat fighting folk) have been trying to shift the global conversation about the internet for a decade now away from "Bandwidth" - towards the real factor that actually matters - reliably low round trip times, especially when the link is actually carrying traffic.
The billions being burned o fiber buildouts is one example - sure, you get more bandwidth but another core and underappreciated fiber feature is the vastly reduced - potential - round trip times. But unless you ALSO get better routers, the jitter and latency under working conditions can actually get worse!
Hitchhikers was BRILLIANT: "Why are you building a bypass?" - "Because bypasses MUST be built!" - and "The plans have been on file at alpha centuri for 50 years, I don't understand what you humans are so bothered about"... boom! earth destroyed. For no good reason. Similarly, when we'd made the breakthrough in queue theory that fq_codel was, fixing a problem that the internet had had since 1983, and anti-patented it and showed that few people actually needed more that 20Mbits of bandwidth so long as there was good queue management... and released the source code for free, we'd mistakenly thought the world would just do a little zero cost software upgrade and everybody, even and especially DSL folk, have better internet, all the time!
Perhaps this is the year some world leader or businessman would use bufferbloat coherently in a sentence.
On my guitar is a sticker that says THIS MACHINE KILLS VOGONS - which is a reference to the guide, and the American folksinger who wrote "This land is your land" - a song tied up in a copyright dispute.
Anyway, there's a ton of useful links in more formerly banned post above and I hope y'all can have a laugh at musk's expense. I figured they only had to fix 3 routers, by a recompile...
I was also in my youth very inspired by the honesty and directness of British punk, notably the Stranglers and Killing Joke. One of my best friends, based in bristol, Paul Whitrow died last year, not from covid, but apoplexy. I keep hoping for more Arthur Dents to ask difficult questions of regulators, and more folk "Get in there" to fix their internets for themselves, as we are all stuck, still, working from home.
A huge irony of the need for better videoconferencing is that videoconferencing is actually low bandwidth, it just needs reliably low jitter and queuing delay. Why can't the ISPs or regulators talk about that?
Because bandwidth must be built!
#FreeDave
Last edited:
There's some graphic pictures of how much better you can make DSL on this link here: http://www.taht.net/~d/broadcom_aug9_2018.pdf
(this is a test if I can say "graphic pictures" and post a link now)
(this is a test if I can say "graphic pictures" and post a link now)
And lastly, this is an effort I've been trying to get off the ground in the USA and in India. https://docs.google.com/document/d/1T21on7g1MqQZoK91epUdxLYFGdtyLRgBat0VXoC9e3I/edit comments welcomed.
Nice to see you back, and that we managed to #FreeDave 
There's a lot to wade through, which is always a good thing. It's frustrating that vendors won't implement such simple changes, especially when - as you said - it's a few lines of free code and gives immediate and tangible returns for zero TCO. I had a good laugh at your Elon anecdote/video, it's often the case that those with the power seldom listen to those who know (unless they're very smart). I hope to see you pop back here one weekend for updates/anecdotes/errata, but if not I have your email address should anything come up (not to mention the buffer bloat mailing list). Stay funky...
There's a lot to wade through, which is always a good thing. It's frustrating that vendors won't implement such simple changes, especially when - as you said - it's a few lines of free code and gives immediate and tangible returns for zero TCO. I had a good laugh at your Elon anecdote/video, it's often the case that those with the power seldom listen to those who know (unless they're very smart). I hope to see you pop back here one weekend for updates/anecdotes/errata, but if not I have your email address should anything come up (not to mention the buffer bloat mailing list). Stay funky... This thread started because I was curious about the current latency under load on VM's products. While we are big fans and developers of the "flent" tool, recently apple released a simpler test for bufferbloat, described here: https://randomneuronsfiring.com/using-apples-rpm-tool/
I'll be back later in the week. Thx for the warm welcome.
#FreeDave
I'll be back later in the week. Thx for the warm welcome.
#FreeDave
I'm running a flent tuning session as we speak. The results so far are... probably worthy of their own thread.
Maybe you can pop back later in the week when I have some real data.
Just ran the tool on a VM 1gig service. A medium responsiveness is not ideal right?
@ChrisD. Can you run on yours out of curiosity?
EDIT: Cleaned up output and @Rainmaker Can we see your output too please as you’ve set configuration on your router to combat bufferbloat?
Last login: Mon Nov 8 18:42:38 on ttys000
uvarvu@Orcus ~ % /usr/bin/networkQuality -v
==== SUMMARY ====
Upload capacity: 44.792 Mbps
Download capacity: 894.068 Mbps
Upload flows: 20
Download flows: 16
Responsiveness: Medium (942 RPM)
Base RTT: 21
Start: 08/11/2021, 18:43:27
End: 08/11/2021, 18:43:42
OS Version: Version 12.1 (Build 21C5021h)
Last edited: