Not logged in
Log in now
Create an account
Subscribe to LWN
Pencil, Pencil, and Pencil
Dividing the Linux desktop
LWN.net Weekly Edition for June 13, 2013
A report from pgCon 2013
Little things that matter in language design
The anycast router at 188.8.131.52 in one direction, the nearest router announcing 2002::/16 in the other direction.
DJB was wrong... even if he was right too.
Posted Jan 27, 2011 1:15 UTC (Thu) by bojan (subscriber, #14302)
[bojan@shrek ~]$ dig +short AAAA ipv6.google.com
[bojan@shrek ~]$ ping6 ipv6.google.com
connect: Network is unreachable
[bojan@shrek ~]$ dig +short www.google.com
[bojan@shrek ~]$ ping -c 1 www.google.com
PING www.l.google.com (184.108.40.206) 56(84) bytes of data.
64 bytes from syd01s01-in-f104.1e100.net (220.127.116.11): icmp_req=1 ttl=55 time=18.3 ms
--- www.l.google.com ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms
rtt min/avg/max/mdev = 18.333/18.333/18.333/0.000 ms
As I said, useless.
Posted Jan 27, 2011 1:40 UTC (Thu) by nybble41 (subscriber, #55106)
DJB's proposal to make all public IPv4 addresses directly routable as IPv6 addresses would save a trivial amount of reconfiguration (the easy part) in the short term, while IPv4 addresses are still dominant. It would not avoid the necessity of updating all existing OS and application software, routing hardware, and address-aware network protocols, scripts, etc. to deal with the longer addresses, which is the real bottleneck standing in the way of IPv6 adoption.
Posted Jan 27, 2011 1:50 UTC (Thu) by bojan (subscriber, #14302)
Why should I? I already have a perfectly good setup here. I can browse, e-mail and do other things on the real net just fine. For instance, I can post this comment on LWN just fine :-)
> trivial amount of reconfiguration (the easy part)
Trivial? I think you haven't read the article above:
> To make that transition, we'll have to do more than assign IPv6 addresses to systems. This technology will have to be deployed across something like 1.8 billion people, hundreds of millions of routers, and more. There's lots of fun system administration work to be done; think about all of the firewall configuration scripts which need to be rewritten. Geoff's question to the audience was clear: "you've got 200 days to get this done - what are you doing here??"
Posted Jan 27, 2011 2:33 UTC (Thu) by cesarb (subscriber, #6266)
From my own experience at work, it was very easy - a few lines in /etc/network/interfaces on the firewall/router, and install and configure radvd on the same machine (its configuration was also only a few lines and very simple). That was enough to make every computer on the network have working IPv6 (even the printer, which surprised me). Yes, I did not have to do anything on any machine other than the router.
And there is also teredo. With it, any machine with an IPv4 address, even behind most NAT setups (as long as the teredo port is not blocked by a firewall), can use IPv6 (not only outgoing but also incoming).
Posted Jan 27, 2011 2:46 UTC (Thu) by bojan (subscriber, #14302)
See the sentence in my post you replied to:
> No, his idea is a common sense one: people already connected to the net should not have to reconnect.
I have no idea why people are trying to justify this screwed up plan. It's a bit like Itanic. We'll have this beautiful new CPU, but _all_ application writers will have to optimise software for it when they compile their apps. Brilliant plan. Just like the current reconfiguration of every single working setup out there.
Posted Jan 27, 2011 3:21 UTC (Thu) by foom (subscriber, #14868)
Sorry, but that's bullshit. Joe Sixpack's computer will *automatically configure itself* to start using IPv6 as soon as it's attached to a network which supports it, similarly to how it automatically configures itself to use IPv4.
Mr. Fancypants System Administrator might have to change *his* configuration to get all his services talking IPv6 (he probably has custom firewall rules, has services listening on specified some subset of IP addresses assigned to the host, etcetc), but Joe Sixpack most certainly does not -- his OS just supports it out of the box, and will magically start working.
Getting everyone using routers with IPv6 support is the tricky part. But that's required no matter what!
Posted Jan 27, 2011 3:30 UTC (Thu) by bojan (subscriber, #14302)
Oh, really? Right now we have lots of such networks floating around? 20 years after the problem has been identified and we're on the brink of IPv4 address space exhaustion? Not even close. That exactly is the problem.
> Mr. Fancypants System Administrator might have to change *his* configuration to get all his services talking IPv6 (he probably has custom firewall rules, has services listening on specified some subset of IP addresses assigned to the host, etcetc), but Joe Sixpack most certainly does not -- his OS just supports it out of the box, and will magically start working.
So, both are really screwed. Joe Sixpack has no network to connect to (e.g. me being Joe in this case - no IPv6 in sight unless I specifically buy new equipment, request IPv6 from my ISP etc.). And Mr. Fancypants is up for many a late night. Just peachy.
> Getting everyone using routers with IPv6 support is the tricky part. But that's required no matter what!
Well, yeah. If the plan to have v4 embedded into v6 was followed, this would have been around years ago. And you wouldn't even know you had it.
Posted Jan 27, 2011 8:16 UTC (Thu) by roblucid (subscriber, #48964)
What's needed is a global IPv6 network. That requires the big content providers and the big ISPs to meet and say, "we want to be able to serve new customers not stagnate" so we need to invest and facilitate a transition.
That will only happen, and the investment made when the address scarcity bites.
Posted Jan 27, 2011 8:31 UTC (Thu) by bronson (subscriber, #4806)
And maybe not even then. There's still a good chance that someone will invent a more practical solution than ipv6. Necessity can be such wonderful inspiration.
Posted Jan 27, 2011 10:10 UTC (Thu) by dan_a (subscriber, #5325)
By aggregating together large numbers of fragmented IPv4 routes which belong to the same autonomous system the routing table shrinks massively which theoretically reduces router memory requirements and routing table lookup times.
If we kept the old, fragmented, IPv4 routes in the new IPv6 table then this benefit would never ever be felt. That's why the global migration plan does not include embedding existing IPv4 addresses in IPv6.
The routing table issue
Posted Jan 27, 2011 14:48 UTC (Thu) by mstefani (subscriber, #31644)
IPv4 addresses run out this year? We run into the routing table issue 8 years ago. Trying to route something smaller than a /24 is an exercise in futility. But we had even fun like "Your IP address is from an IP space that is allocated in /20 blocks; we are filtering out any routes smaller than that. Have a nice day."
Posted Jan 27, 2011 15:41 UTC (Thu) by foom (subscriber, #14868)
Actually the (completely unworkable) idea was that people who had multiple providers would simply have multiple IP addresses, and advertise all of them. That *possibly* could have even worked (with a lot of effort) if TCP supported multiple endpoints and transparently switched between them in real-time. But, it doesn't.
The one (pretty minor) remaining thing IPv6 does to help reduce routing-table size is reduce fragmentation of the address space -- a single organization at a single location is less likely to need multiple non-contiguous addresses spaces than in IPv4.
Posted Jan 28, 2011 7:20 UTC (Fri) by butlerm (subscriber, #13312)
That is why they invented SCTP, which does all that and more. Perhaps too much even.
Posted Jan 28, 2011 7:18 UTC (Fri) by cmccabe (guest, #60281)
On the other hand, the world population growth rate is well under 1% per year (according to wikipedia.)
Sound like the routing table problem will solve itself pretty quickly, without us doing a thing.
Posted Jan 28, 2011 12:20 UTC (Fri) by khim (subscriber, #9252)
Posted Jan 27, 2011 13:49 UTC (Thu) by foom (subscriber, #14868)
So why would anyone have added support for an IPv4 "long address" option which wasn't even being used by anyone? They wouldn't have, any more than they added IPv6 support. Both things are equivalent: addition of a new feature to your software that none of your customers are going to actually use.
My router could have added IPv6 support 10 years ago, too, without me knowing it. But it didn't.
Cable modems didn't get IPv6 support until DOCSIS 3.0, which only *just* started showing up in new devices. It looks like they've also released an addendum to DOCSIS 2.0 to allow older hardware to add IPv6 support -- in November 2010. So *maybe* firmware upgrades for older devices might add IPv6 support in the future (but I'm not holding my breath)...
Posted Jan 27, 2011 20:08 UTC (Thu) by dlang (✭ supporter ✭, #313)
the 'implement a new stack, with mandatory features that are complex, and largely unknown' aspect of IPv6 made it a major project, not something that could be added in easily.
Posted Jan 27, 2011 21:15 UTC (Thu) by tialaramex (subscriber, #21167)
So software is great, it's enough to have IPv6 in your house, share it with a few friends in a tiny startup's office, or even across a few hundred hosts if you can afford a beefy FreeBSD box as a router. All these things were being done last _century_ in preparation for the transition we are now undertaking.
But that only gets you so far. Eventually (today probably somewhere about gigabit speed) it is only cost effective to switch IP with custom hardware designed for that specific purpose. Eventually the cost of wider addresses doesn't vanish, but instead dominates. And there the story changes.
Posted Jan 27, 2011 21:31 UTC (Thu) by dlang (✭ supporter ✭, #313)
this is part of the problem, by designing IPv6 so that everything had to change we have these problems. If this was something backwards compatible, that could run through existing devices in the middle without them having to change (think of it as being similar to tunneling, but with every ISP being a potential tunnel termination point without having to configure it explicitly) then we would not have to deal with the ISPs support or lack of it as an issue, only the question of getting enough endpoints to support it.
this wouldn't have solved the problem entirely, but it would have helped.
And if you are claiming that all developers who matter have implemented IPv6, please go back to the earlier post that pointed out that no current generation game console supports IPv6, many printers don't support IPv6, let alone other, smaller embedded systems.
Posted Jan 28, 2011 16:13 UTC (Fri) by tialaramex (subscriber, #21167)
The ISPs don't provide such tunnels because the tunnel would _cost money_ and they don't want to spend money.
Posted Jan 27, 2011 22:09 UTC (Thu) by bojan (subscriber, #14302)
Equivalent? I can't even test IPv6 without obtaining a new address. I could if my IPv4 worked instead. That's why it would get included - it would be actually useful at some point to people _connected_ to the net. In fact, we wouldn't even be talking about the address crunch.
> My router could have added IPv6 support 10 years ago, too, without me knowing it. But it didn't.
Yeah, of course it didn't. Where would you get an IPv6 address from anyway?
Posted Jan 28, 2011 15:09 UTC (Fri) by zlynx (subscriber, #2285)
Did you miss that every IPv4 address is automatically assigned an entire IPv6 network of its own in the 2002::0 range?
Here is mine, for example: 2002:4051:69fa::1 You should be able to get the web page at http://[2002:4051:69fa::1] or http://oberon.zlynx.org/ The DNS has both A and AAAA records.
See the 4051:69fa? That's my IPv4 address: 18.104.22.168
If home routers were preconfigured with 6to4 and radvd, every home user would already be on IPv6.
Sure, an inefficient IPv6 that is tunneled through IPv4, but it sets the stage for moving to ISP assigned addresses later on.
Posted Jan 28, 2011 15:53 UTC (Fri) by tialaramex (subscriber, #21167)
In this scenario you are an IPv6 island, connected to other IPv6 islands by tunnel. If they _wanted to_ ISPs could compete by providing a tunnel endpoint closer to their subscribers, so IPv6 would work faster on their network than on a competitors. Technologies like 6to4 even made this very easy by using an anycast address, so the ISP can put the tunnel endpoint anywhere with no config change for the user. The natural endpoint of such competition would be native IPv6, no tunnels.
But the major ISPs as we've tried repeatedly to explain, do not care. If the closest endpoint is on another continent, why should they care? Providing the minimum possible service is a cost saving. So if you try to use such a tunnel, expect poor performance and zero technical support from your ISP. It's just not in their interest to care.
Have I told the story about the cable TV company who added a "digital surcharge" to pay for equipment upgrades ready for digital cable? I bet you're thinking that they had to do that - to pay for the upgrades, right? Nope, there were no upgrades. It was just another way to get more money. When digital cable actually arrived they couldn't offer it, because their equipment was too old. No matter, the customers still have money, just charge them again.
Copyright © 2013, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds