| |||||||||||||
The CoDel queue management algorithmThe CoDel queue management algorithmPosted May 28, 2012 18:13 UTC (Mon) by Jonno (subscriber, #49613)In reply to: The CoDel queue management algorithm by nye Parent article: The CoDel queue management algorithm
I've got a pretty standard 100/10 mbps residential internet connection (at $23/month), and my ping times are not nearly that high.
First hop (actually two hops, as it goes through my Linux router):
To server on national university backbone 270 km away:
To residential ADSL user with different ISP over an OpenVPN network:
(Log in to post comments)
The CoDel queue management algorithm Posted May 28, 2012 22:47 UTC (Mon) by raven667 (subscriber, #5198) [Link]
I wouldn't say that's standard, 100Mbit is way at the top end of the range at $100/mo+, if it is available at all. 10/1 or 4/0.5 are much more common and only cost $30-$50. You must have some incredible infrastructure where you are at. Also my numbers match up with the poster, there is 10-20 ms of latency between my laptop and my first upstream hop. Maybe you are on fiber and don't know it?
The CoDel queue management algorithm Posted May 29, 2012 1:59 UTC (Tue) by dlang (✭ supporter ✭, #313) [Link]
there are still a LOT of people stuck with 1.5 or 3Mb down and 768Mb up. This is worlds better than dialup, but if developers are considering multiMb in each direction 'typical', it's no wonder that they think that having everything in the cloud will work (or that doing network OS installs should be the norm), and then wonder why they have so few users.
The CoDel queue management algorithm Posted May 29, 2012 6:31 UTC (Tue) by Jonno (subscriber, #49613) [Link]
I've got an ethernet jack in the hallway labelled "Internet", which is quite typical in Swedish apartment complexes. 100/10 Mbps only costs $23/month, though a business line with bandwidth guarantees cost much more (I usually only get 60-80 Mbps down and 8-9 Mbps up, and occasionally only 30-40 Mbps down).
People living in houses rather than apartments are usually stuck with 20/1 Mbps ADSL, I know my dad pays $17/month for that speed, and his ping times are worse than mine. But when I ping him over our VPN tunnel it is still way below the 30-40 ms you quoted as a minimum.
While I know that some backwater nations, such as the US, have all but no working infrastructure, don't assume that is the norm, that is the exception.
The CoDel queue management algorithm Posted May 29, 2012 7:46 UTC (Tue) by spaetz (subscriber, #32870) [Link]
> While I know that some backwater nations, such as the US, have all but no working infrastructure, don't assume that is the norm
Let me assure you that this is *not* the norm even in EU countries (can't speak for Asia). If you are below 20-30ms ping time for the first hop, you are the exception. (perhaps not in Sweden, but in Europe for sure).
And there is no need to insult countries here.
The CoDel queue management algorithm Posted May 29, 2012 8:47 UTC (Tue) by dlang (✭ supporter ✭, #313) [Link]
the simple truth is that it's far easier to wire a small country than it is to wire a large county.
If you have a high population density you also have the ability to spread the cost of doing so across many more people.
and frankly, it helps to be late to the party as you only have to implement the latest and best technology, not each generation as it is developed (never mind paying the cost of the development)
the result is that in small, high population density countries you can have really good infrastructure, but in larger areas you aren't going to have nearly as goon an infrastructure, not matter what the cost.
I live in the greater Los Angeles area, but out around the edge of it. I pay $130/month for 1.5Mb down/768Kb up. I could upgrade to an ethernet connection up to 5Mb, but then it would cost me $100 per Mb.
and I'm in a 'good' (but not 'great') area for connectivity. Where my sister lives, their only option is satellite (with a ~1000 ms first hop ping time) to give you an idea of how remote they are, they literally live 20 miles from the nearest fast food place.
now there are places in the US with connectivity almost as good as what you get (although at higher out-of-pocket costs, I'm assuming that your system has some tax money included in it), and I wouldn't lay a bet either way on if there is more area covered with such good coverage in the US vs in Sweden, but it could be several multiples larger in the US and not make a significant dent in the problem.
The CoDel queue management algorithm Posted May 29, 2012 13:46 UTC (Tue) by Jonno (subscriber, #49613) [Link]
> the simple truth is that it's far easier to wire a small country than it
> is to wire a large county.
Sweden may only be 5% of the US, but it's still larger than any US state but Texas and Alaska. In a more fair country-to-country comparison, Sweden is larger than for example Germany, Italy and the United Kingdoms.
> If you have a high population density you also have the ability to
Well, Sweden has 20.6 residents/km², while US has 33.7 residents/km², so obviously US should have much better Internet connection than Sweden...
> and frankly, it helps to be late to the party as you only have to
Well, the Swedish broadband infrastructure project began back in 1998, and for the last few years people have started to complain that next-to-nothing have happened for over 5 years. While most of Sweden have been upgraded to ethernet, fiber or cable connections over the years, 35% of all Swedish households are still limited to the previous generation broadband access, because the goverment stopped subsidizing broadband infrastructure projects once ADSL was deployed everywhere (well, 99.91% of households) back in 2005.
Yes, ADSL is considered previous generation broadband access in Sweden, even though base stations have been upgraded to support 24/1 Mbps compared to the 8/1 Mbps that was common in 2005.
The current generation of broadband access (usually an ethernet jack, sometimes a cable-tv modem) started to roll out in 2000, but wasn't common until 2005, when it reached 40% coverage. Today that figure is 65%. Personally, I got a "real" broadband connection in 2002, albeit back then the speed was only 10/2 Mbps and it cost $28 per month. I got my current 100/10 Mbps connection when I moved in 2004, though at the time it was quite expensive at $45 per month.
> I live in the greater Los Angeles area, but out around the edge of it.
Poor soul, for that kind of money ($126) I could get 250/100 Mbps. Of course, that is because I live in an apartment complex in the middle of a medium-sized town. Most rural residents can't get anything better than ADSL at 24/1 Mbps, and will have to pay $49 to get even that much.
> I'm assuming that your system has some tax money included in it
Well, 2002 through 2005 the government subsidized about half the cost of all broadband infrastructure project, but since then they have only subsidized rural broadband projects, and usually only in the form of targeted low interest loans.
The CoDel queue management algorithm Posted May 30, 2012 13:50 UTC (Wed) by job (guest, #670) [Link]
Higher population density? Late to the party? Please try to get your facts straight next time if you're going to waste electrons on a post.
The CoDel queue management algorithm Posted May 30, 2012 14:13 UTC (Wed) by Cyberax (✭ supporter ✭, #52523) [Link]
In Ukraine no tax money go into ISP business. Indeed, competition is fierce.
Yet we get 100Mb for $8 a month in cities.
The CoDel queue management algorithm Posted May 29, 2012 22:05 UTC (Tue) by nix (subscriber, #2304) [Link]
Quite. I *might* be able to get this in the UK with fibre to the premises (rare as hen's teeth), but fibre to the cabinet couldn't do it and my existing bonded ADSL connection to an exchange <100m away has no hope:
1 spindle.srvr.nix (192.168.16.15) 0.056 ms 0.040 ms 0.063 ms
(as usual, the timing figures from the last two hops are upper bounds because those machines are going to be delaying ICMP replies as they see fit: still, it's in the 20ms range, not the 3ms range).
The CoDel queue management algorithm Posted May 30, 2012 10:08 UTC (Wed) by nye (guest, #51576) [Link]
>my existing bonded ADSL connection
Now we're *really* OT, but what are you using to do the bonding?
I was hoping to do that with a couple of Sangoma PCI ADSL2 modems, but in the small print it turned out that you can only do bonding with them if you're using PPPoE, and I don't know of any UK ISPs that offer that.
I've seen ADSL cards that can do it, but at that point you're better of just going to ADLS2 unless you're doing 4-way bonding or more, and I've seen expensive proprietary boxes that just sort it all out for you, but nothing affordable for an individual or small business.
The CoDel queue management algorithm Posted May 30, 2012 10:46 UTC (Wed) by TomH (subscriber, #56149) [Link]
Well http://compton.nu/2009/12/per-packet-load-balancing-with-... explains how I bonded four ADSL links with the same ISP.
We're bonding two fibre to the cabinet lines instead now, which use PPPoE and makes things a bit simpler as they are terminated with pppd on the linux box where they can be bonded with teql.
In fact the BT Wholesale ADSL platform does support PPPoE but most ISPs don't tell you that... I did have my line at home running that way for a short while though, in preparation for upgrading it to fibre to the cabinet.
The CoDel queue management algorithm Posted May 30, 2012 14:32 UTC (Wed) by nye (guest, #51576) [Link]
>Well http://compton.nu/2009/12/per-packet-load-balancing-with-... explains how I bonded four ADSL links with the same ISP.
Thank you for that link. I hope nobody minds the thread-hijacking to ask a little more about this - it's not a topic that seems to have much Google juice.
With that setup, do I understand correctly that - assuming the ISP has the bonding set up on their end - configuring the TEQL interface as you have will work regardless of the connection method used? It looks like it's actually very simple so long as you know the secret sauce.
What I'm not sure about is the IP addresses on the routers. You say that they need to support two different *LAN* addresses; how are the WAN ports configured? Are they bridged or do they need an additional address?
Say for the sake of example that you have a /29 netblock, giving you 6 addresses once you've accounted for the network and broadcast addresses. You use one that's shared between the routers and one for the teql0 interface, leaving 4 available for other machines. Is that correct?
Incidentally, would you recommend the Zyxel P-660 series? I do wonder if a better modem/router might solve my periodic loss of ADSL synchronisation, and while I've tried two different models, it's entirely plausible that they're both fairly intolerant of noisy lines.
The CoDel queue management algorithm Posted May 30, 2012 14:50 UTC (Wed) by TomH (subscriber, #56149) [Link]
Those routers allow the LAN interface to be given up to two aliases in addition to their primary address.
So what I did was to set the primary address to a unique RFC1918 address, which was just used for management purposes when I needed to telnet to a specific router, and then to set the alias on each router to the same, shared, public address.
I then added static routes on each router to pass traffic for our public IPs back to the linux box where the bonding was done.
The WAN ports were configured with separate public addresses - a unique one for each router.
Bridge mode wasn't used - they were acting as normal routers.
I wouldn't like to say if the P660 is particularly good or bad - they were the free routers our ISP provided with the lines.
The CoDel queue management algorithm Posted May 30, 2012 15:16 UTC (Wed) by nye (guest, #51576) [Link]
>The WAN ports were configured with separate public addresses - a unique one for each router.
Interesting, so this sounds like a different configuration than some bonding setups which appear to require only a single public IP address. I wonder if that's down to how the ISP configures their interfaces.
I suspect I could make a lot more progress here if experimentation didn't mean scheduling connection downtime of an unknown duration, which in practice means being physically present in a locked office building in the dead of night (for which honestly they Do Not Pay Me Enough™).
At any rate, thanks for the information.
The CoDel queue management algorithm Posted May 31, 2012 11:21 UTC (Thu) by nix (subscriber, #2304) [Link]
The question really is, will any respondents to this question be using any ISP *other* than AAISP? They really do seem to be the only people who've tried to make line bonding work in any meaningful way, in the UK at least.
The CoDel queue management algorithm Posted May 31, 2012 13:26 UTC (Thu) by james (subscriber, #1325) [Link] This is going somewhat off-topic, but Eclipse do, although I haven't used it.
The CoDel queue management algorithm Posted May 31, 2012 15:03 UTC (Thu) by nye (guest, #51576) [Link]
Entanet also offer a bonded option, and they're one of the wholesalers used by UKFSN.
The CoDel queue management algorithm Posted May 31, 2012 11:20 UTC (Thu) by nix (subscriber, #2304) [Link]
All I'm using to bond is a multihop default route
ip route add default nexthop via 81.187.191.130 dev adsl weight 1 nexthop via 81.187.191.132 dev bdsl weight 1
to Zyxel P-660R ADSL routers that have no idea I'm bonding them, and two of Julian Anastasov's patches from http://www.ssi.bg/~ja/, to wit 00_static_routes-2.6.39-15.diff and 01_alt_routes-3.0-12.diff, to ensure that when one hop's routes go stale because of upstream problems we switch to the other. (I can't rely on normal 'when the link goes dead' bonding-driver stuff because the link that would go dead comes out of the router. I could fix this by using bridging, but Zyxel's documentation for that setup is so appalling that I haven't tried to make that work yet.)
However, I am lucky in that my ISP (AAISP) provides direct support for bonding on the ISP end: any packets sent to my public IPv4 or IPv6 address ranges will end up being evenly scattered between my lines (which fortunately are of similar speed, see the weight above). So all I have to handle is outbound routing.
The CoDel queue management algorithm Posted May 29, 2012 9:13 UTC (Tue) by Cyberax (✭ supporter ✭, #52523) [Link]
>cyberax@cybnb:~$ traceroute lwsrv
>traceroute to lwsrv (94.45.58.146), 30 hops max, 60 byte packets > 1 192.168.18.100 (192.168.18.100) 3.236 ms 3.237 ms 3.247 ms > 2 unallocated.sta.lan.com.ua (92.249.102.1) 3.256 ms 3.263 ms 3.272 ms > 3 out.ua-ix.lan.com.ua (92.249.120.249) 45.030 ms 45.046 ms 45.021 ms > 4 dtel-ix.as6723.net (193.25.180.61) 3.197 ms 5.642 ms 5.714 ms > 5 j2-ua.gw.skif.com.ua (91.90.19.210) 6.323 ms 6.332 ms 8.713 ms > 6 something.com.ua (94.45.x.x) 3.012 ms 5.023 ms 17.928 ms
That's a trace from my residential 100Mb to my office (some 30km away). And I pay $6 a month for it.
You Americans are stuck in a stone age :)
The CoDel queue management algorithm Posted May 29, 2012 15:06 UTC (Tue) by nye (guest, #51576) [Link]
Well I'm from the UK, so my national bias is to assume that 'residential' == 'ADSL', where I don't believe it's technically possible to get pings that low. In some places you can get cable (with a choice of Virgin Media or... Virgin Media) and while I don't know so much about that, by reputation the ping times are at least as bad, often worse.
Over the last couple of years we've been seeing a major upgrade rollout, which means that people in urban areas have a fairly good chance of getting VDSL2, though there's only one wholesale provider and IIUC everyone offering it is reselling exactly the same product, which has fairly low caps until you start paying crazy money (eg £25-30/month for a 30GB limit, or ~£75 if you go all the way up to 180GB).
I don't actually know if VDSL2 is any better in terms of latency though, quite possibly not. Personally I'm relatively happy with my un-capped ~10Mb ADSL2 connection for £18/month, that usually only drops out for a couple of minutes at a time a few times per day (emphasis on 'relatively'). The national average is far worse than that, but we're only about 450m from the exchange.
Do you know what technology you're using?
The CoDel queue management algorithm Posted May 29, 2012 15:23 UTC (Tue) by Cyberax (✭ supporter ✭, #52523) [Link]
Sure. I live in apartment building and get Ethernet to my router (actually, to my Linux machine), with simple static IP assignment (no DHCP, PPP, etc). I can also get routed networks up to /28 if I need it for small additional price.
Apartment building is connected by 10G fiber to my provider's hub which is connected to the main national Internet traffic exchange ( http://www.ua-ix.net.ua/eng.phtml ). It's simple and fast, and works great in dense residential areas.
In contrast, I have to pay $50 a month for 30/5 cable connection when I live in Brooklyn in the US. And it's still considered to be pretty cheap :(
The CoDel queue management algorithm Posted Jul 25, 2012 12:13 UTC (Wed) by farnz (guest, #17727) [Link] All DSL technologies in current use run at a 4 kilobaud signalling rate; higher speeds are provided by running more bits per symbol, so your 10MBit/s ADSL2 is 2,500 bits per symbol, while 40M VDSL2 would be 10,000 bits per symbol. This puts a lower bound on DSL latency of 0.25 milliseconds; practically, VDSL2 seems able to reliably reach the 5ms RTT range for the DSL link itself, while my home ADSL2+ has around 25ms RTT.
|
|||||||||||||