An update on IPv6
potaroo.net
potaroo.net
I work in a lot of regions in the world, (Mexico, UAE), where they've been used to the reality that they won't be assigned any IPv4 addresses. I'm talking about large $Billion+ organizations that made it abundantly clear to me, that the applications we were deploying ($100million plus network deployments) - would have to operate with PAT, or we would not be able to deploy them.
And so we worked around the requirements that we deploy everything on PAT, and our network deployments (sometimes with thousands of IPV4 network elements) would have to operate purely within RFC1918 address space.
Speaking as someone who has worked on million+ IPv6 node deployments, (So I clearly don't have a bias against IPv6) I can say with some certainty that the only "calamity" we are going to see is increased use of PAT on clients, and somewhat more expensive IPv4 addresses for those who need to act as a server. Everyone else will be fine. (Have been fine for years in places where public IPv4 addresses ceased to be available).
The reality is that this lack of pressure to transfer to IPv6 means that we're going to see an extended period of time in which we migrate. It will happen eventually (IPv6 has a lot of awesome features, particularly around self-addressing, and lord is it wonderful not to have to play CIDR addressing games, everything is a ::/64) - but it's going to be a lot longer than most IPv6 advocates would prefer.
/64 is the smallest network, but there are certainly larger aggregates with smaller prefixes.
In particular, some ISPs will hand out a /56 - allowing the home user 256 /64s. But, once again, these are administrative routing decisions - on the interface, it's always a /64 or (in some rare cases - though I'd love to hear if people still see this in the wild) /127s for router-router connections.
You will never, ever see, for example, a /72 IPv6 network in the wild.
For those IPv4 network engineers who have a decade+ of jugglings /25s, /26s, /27s, /28s and /29s, and trying to preserve IP space (particularly globally routable), this is nice change of affairs.
And even just talking about "this is global unicast space", "this is 6to4 space" and "this is link local space" involves small prefixes. So, for example, you still need to do the same mental juggling to know that 2000::/3 is 2000... to 3fff....
Thankfully, the 100x more common task of configuring an interface, will always be a /64.
Quite the calamity if you ask me: Users have to actively ask for the permission to act as a server. And at a significant monetary cost, if I understand the constraints correctly. Sure, right now, nobody has a server at home. But if we did (and this is certainly technically possible), then centralised spying-ridden services such as Gmail, Facebook, and YouTube would be mostly useless. (Assuming peer to peer distribution, you don't need a server farm to scale.)
This would also help with routing by the way, by keeping communications more local.
I don't think it's that uncommon (and I say that not only because I have one myself). Home routers make it trivial to add one, and NAS boxes are more common than ever. Almost every game and torrent client needs to have ports forwarded (which it a server in the network sense), and UPnP is a "feature" most people have that makes it possible to be unaware of it all.
A lot of video-games have the host of a match actually work as a server. NAS as servers. In lots of inter-device applications, one of them works as a server. Oh, right, and Xorg.
Right now, many people, mostly in countries that are lacking in IPv4 allocation, do not have an internet connection.
An internet connection is a public IP. If you're behind a NAT, or something like it, you don't have a public IP, therefore you are not connnected to the internet. You are connected to a private network, which forwards some of your packets to the open internet. And that server thing. No public IP, no server. Not such a big problem when you control the NAT (assuming it's in your router), and can forward some ports. But if you share the same public IP with 50 other customers, it is theoretically impossible to have more than one HTTP server for those 50 people. The practical limit, of course, is zero.
I've spent the majority of my working life in the last three years happily accessing the internet, internet services, and I've always been behind a PAT device when in a hotel, serviced apartment, etc...
For example, right now, in my Office in Singapore, my IP address is: 192.168.108.34. And I do all my conferences, whiteboarding, VPNs, etc.. from this office.
Now - back to your scenario - You certainly can share a lot of HTTP Servers on a single IP address - a lot of the customers I've worked with do so using PAT, and, you get very used to the port number being part of the HTTP server address:
In a ultra-low IP address environment, you have teams with not only IPAM (IP address Management tools), but also Port Address Management - they track who has been allocated to which port.
Now - is this the best case scenario? - No, obviously not - it would be awesome to have a Static IP address, and, there are some scenarios in which I have to have a static IP address - in particular getting IPSec VPNs through PAT can be problematic, so if I want to run a VPN concentrator, I start begging and pleading to get a static IP address. But I know this is going to cost me a lot of money. And that's where I started - once we run out of IPv4 addresses, the market will just start bidding the price of them up. Right now, most ISPs in Singapore will give you a static IP address for $120/year. I expect, as they run out of IP addresses, that price will start to increase - and people who might have opted in for one at $120/year, will probably decline at $500/year - returning those IP addresses back into circulation.
At the same time, if someone is sitting on a /20 that they aren't using very using very much (4,000 addresses) - and those IP addresses have $2million/year of value to other people, they will be highly incentivized to capture that value.
With all that said - was it not wonderful to call Layer 42, and, within 30 seconds, be granted 2001:1868:209::/48 off of their /32 ISP range? Yes, it was. I had 65,000 networks I could play with, each of which had effectively infinite address space. Do I look forward to that the day that is universal, in which every ISP user is immediately granted a /64 without asking, and anybody with a home network gets a /56 without paperwork, and a /48 is available with just a quick, inexpensive surcharge? Yes, also looking forward to that.
But - "Disaster" and "Catastrophe" doesn't capture what will happen if IPv4 takes a few extra years to get traction. The market and people's behavior will adjust.
IPv6 deployment is massively late, users have already been conditioned into accepting NAT as internet access, even going so far as NAT boxes being called "routers"- even though NAT flatly contradicts the RFC definition of IP router, which are forbidden from doing port/address NAT along with other arbitrary traffic interference.
One thing that people need to be precise about, particularly as IPv6 comes online, is the difference between NAT and PAT. I realize that 90%+ of the HN community understand the difference, (NAT changes your Address only, PAT changes your address to a single outbound address, and multiplexes a PORT) but with IPv6 it has a very real significance.
Deployment of PAT in IPv6 is really quite frowned upon, but there a number of very real administrative reason why Network Prefix Translation (NPT, https://tools.ietf.org/html/rfc6296 ) - which is really just stateless NAT, makes sense.
With NPT, you convert the first 64 bits (the "Network"), but leave the last 64 bits (the "Interface") untouched. This gives the inside of your network really good ISP mobility - allowing you to switch to a new ISP at the drop of a hat, without having to worry about any internal renumbering, updating thousands of DNS records, etc... The only DNS records you would need to worry about would be for external connectivity (which is typically orders of magnitude less than internal hosts, and also typically more well documented.
I expect that most forward looking enterprises deploying IPv6 will use some combination of RFC 4193 (Unique Local IPV6 Unicast, https://tools.ietf.org/html/rfc4193, the IPv6 variant of RFC 1918) + RFC 6296 Network Prefix Translation.
My point is, it does. It is not an "end of the wolrd" disaster, but it is an Orwellian one. We're only beginning to realise the insanity behind centralised services such as Gmail. I personally suspected those dangers for a long time, (as did the FSF, the EFF, and many others). Edward Snowden only confirmed some of our worst fears.
That said, the lack of public IP is only half of the catastrophe. The other half is asymmetric bandwidth.
So IPv6 only connections will start to become a vialbe end-user product. These will be cheaper than dual-stack installations, so the ISPs will push them.
This isn't really necessarily the case. The full details on the requirements for Apple's required "support" for IPv6 are sparse still, but there's no reason that this can't be done with NAT64.
You could even have made IPv4 addresses variable length this way, from 32 up to a very large number of bits, and hierarchical. And it's relatively easy to make such a scheme backwards compatible, because you can have (say) a webserver sitting on 1.2.3.4 answering queries for older clients that don't know how to form a 1.2.3.4.5.6 address packet.
The opportunity to break backwards compatibility in v6 brought many other good things as well: fixed fields that are easier to route, more sane network allocation schemes, router discovery for LANs, and lots of other bits and pieces.
Because that wouldn't be compatible with existing IPv4 deployments and would cause a reliability havoc when a node not configured to deal with it mangled packets transparently somewhere in between your source and destination end-points.
I don't get the resistance against IPv6. It works. It's a fresh take. Yes it requires some new stuff to be deployed and configured here and there, maybe even requires you to learn something new, but if you thought extending IPv4 would have been any other way you are deluding yourself.
If you're going to do a significant change to something as big as the internet (and introducing a new address scheme is that, no matter how you implement it), you might as well step back and think it all through instead of applying yet another hack.
So tell me. Why are you opposed to IPv6? Why do you want to hang on to this old IPv4-thing which is already at the bursting point, at the edge of what it can take?
I'm very concerned about implications for anonymity. If every device has a unique IPv6 address, then just one leak can compromise all other anonymized connections.
IP isn't anonymous; it wasn't designed to be, and probably (given the problems it's trying to solve) shouldn't be. Anonymity should be an overlay.
I totally agree, though, anonymity is not part of IP. I was not implying that, just that there are still ways for home users to somehow shuffle their IPs within the IPv6 framework.
I get that the Tor Project is working on IPv6. But I also want IPv6 NAT, in pfSense or whatever. That's to keep my local device IPv6 addresses (hosts and VMs) private, even from Tor entry guards. I guess that it's time to learn how to ensure that.
IPv4 did not offer any anonymity. The closest you get to that, is an external entity being unable to differentiate the traffic you make vs the one your family makes.
http://www.spinics.net/linux/fedora/fedora-users/msg459370.h... (tldr: It's 2015, yet the default firewall in a major Linux distro can't handle IPv6)
Could you explain what this means?
But do you have a way to discover whether any of the first hundred devices break soon? Or even the 150 new devices? You could add a device on a new-only address before you run completely out of the old-compatible addresses, but that won't test completely. There might still be devices that will connect, but log the connections improperly, or have a buffer overrun and crash randomly, or, or, or. There might be software on a device that supports only old addresses, even though the device itself supports the extension scheme.
If the network is a big one, where different parties own and operate the devices, then you may well be completely unable to test whether the extension scheme is universally supported. Before the day when you run completely out of unextended addresses, that is.
Contrast this with a dualstack approach, where you can make a list at any time: THESE devices are on v4 only and will be a problem when v4 runs out, THOSE OTHERS are on v4+v6.
One of the many studies: http://www.eecs.berkeley.edu/Pubs/TechRpts/2005/EECS-2005-24...
Any solution will involve some changes to network infrastructure.
If they had simply added some bytes to the address and not tried to tackle every other perceived wrong with IPv4, we wouldn't be talking about this exact same problem 17 years after IPv6 was introduced.
I personally believe the biggest driver for IPv6 adoption will be Apple requiring support in iOS apps. Right now, IPv4 space is still comparatively "cheap"
This translates into super-large ramifications, especially for companies like Comcast--I can only imagine that they don't want to add yet another possible moving part, especially when there's almost zero benefit to the average customer.
It's the smaller ISPs that are just used to fixing stuff "when its broken" that are really in trouble.
What does surprise me though is that IPv6 hasn't seen steady uptake when new companies set up new infrastructure from scratch or existing companies upgrade their kit. I would have thought that using IPv6 exclusively internally with DNS64/NAT64 to bridge the external IPv4 internet would offer some advantages.
From what I've heard from cable companies is Cisco. They've had many issues with Cisco interop IPv6 on HCF networks.
Here's an enterprise perspective:
It's IP address management. it's massive reconfiguration of every node and network endpoint. It's auditing and completely redeploying security measures that are network based. It's training every IT employee on network basics from the ground up.
It's millions upon millions of dollars in man-hours and infrastructure and tooling for something that might have marginal benefit on an internal network. I agree the new world will be better with larger subnets and many of the features of IPv6, but it's more complex.
Fiber-optic and hardware level upgrades touch a lot fewer devices, and are transparent to users and applications.
In fact, I would suggest that NAT is responsible for a lot of that centralization, as the workaround for not being able to listen(2) for your friend you want to communicate with is a centralized service to manage that connection.
One of the original benefits of TCP/IP networking was that every peer was equally client and server in the eyes of the protocol. The damage done by NAT has been the de facto removal of that benefit. Even now when reasonably-reliable (but INCREDIBLY complicated) NAT-traversal protocols exist, a centralized service is still required, nullifying the benefit of being able to publish without the permission[1] of a central authority.
Wasn't that exactly GP's point?
They are mutually reinforcing each other. Unfortunately, ISPs would only act on consumer pressure, and consumer pressure won't exist because they don't know of all the other nonexisting software that could exists without NAT. We have centralized internet because of NAT, and we have NAT because centralized services are dominant.
IMHO, IPv6 won't fix this problem for many cases. Consumer routers still won't allow inbound connections, so there will still be the need for some kind of way to allow the client to tell the router to forward traffic on a specific port (likely https://en.wikipedia.org/wiki/Port_Control_Protocol).
Which is exactly the same as we get these days with NAP-PMP or UPnP.
Maybe once listening on all local interfaces will not listen on temporary addresses, we can move to a one-address-per-service model at which point it becomes more feasible to forego the central firewall dropping all incoming connections as port-scanning the machine you just got a connection from will involve scanning 65536 * 2^64 ports
As it stands now, I'm not willing to expose all open ports of my internal machines over IPv6. Instead I still have specific port opening rules on my central firewall.
While I have no issue exposing, say, Skype on some port, I certainly don't want to expose, say, my local NTP port of my Mac, or, worse some local port mapper needed for NFS mounts.
Yes. I could use a local firewall, but these by default only allow a per-application setup. "Do you want to allow NTP to be reachable?" - To my local network? Yeah. Maybe. To the internet? Hell no.
Current simple built-in firewalls don't allow this kind of answer - at least on the mac.
A particularly bad misconception that ubiquitous NAT has encouraged is the conflating of the address-translation step (NAT) with the firewall, when the two concepts are not related at all. You can run either independently of the other[1].
All that is needed for a home router is a simple firewall, and a way to open holes in the router's firewall from the internal hosts. The fact that this software isn't already available on most OS/home-routers is simple lack of demand... and the fact that it would be totally useless for current NATed home networks. You don't need all of UPNP or NAT-PMP; there are no ports that need to be mapped. The difficulty in introducing something to allow certain types of packets through a firewall is trivial compared to the complexity of working around NAT.
Concerns about port scanning and "expose all open ports" are just nonsense, and has nothing to do with NAT. This all-or-nothing approach you seem to be assuming misses the entire point of having an external firewall (such as on an home router). You get to pick which packets you let through.
[1] This is why NAT doesn't provide security on it's own. Static NAT, for example, is just substituting the address; all packets to the public address are forwarded to the translated address.
If so, then that could mean that packet sizes are going to be constant over the next 20 years, all the while that protocols and pipes get many times faster and bigger. If I'm not mistaken, those smaller and smaller packet sizes (relative to faster and faster networks) are going to mean tremendous packet switching overhead (routing, encryption) especially in protocols such as QUIC which try to align encryption with packets.
Sort of like a plane having to stop at every red light in a traffic grid designed for cars.
On a single machine. There's still a bit of room for faster connectivity now that everything's hooked up to the internet and streaming ever-bigger HD video. Infrastructure devices might need a bit of an upgrade though when we hit their limits
RFCs should, and at least with the popular ones, generally do, reflect the status of how things actually work on the internet, not how we might wish they did. If your standards ignore reality, they in turn will be ignored.
Rightly or wrongly a LOT of networks block ICMP wholesale, and thus block pmtud, it does not matter if some word in a text document changes somewhere, this will continue to be a reality, so you better be aware of it and be able to cope with it.
Rough consensus and running code.
SHOULDs exist precisely because of situations like pmtud, you cannot gaurentee it will be supported, but when it is supported it is helpful and useful.
Furthermore, no one said this was about "cudgels" and "bludgeons" or being "too strongly legalistic" or writing "laws". Those are your words, not mine.
They should do it on a World IPv6 Day, how about 6/6/16?
Looking at http://www.google.com/intl/en/ipv6/statistics.html it's doubling every year. So it's going to take another four to reach everywhere.
Maybe 6/6/2020 :-)
(And there are plenty of IPv4-only devices out there, like printers, Consoles, etc. So I suspect another five years after that before there's so little left that the address space starts clearing up again.)
Very interesting point of view imho ("bring IPv6 to the rich first, the others will follow").
I still don't know how to use IPV6. Maybe it is just too complicated and they should come up with something simpler?
128 bits of address space. Everyone gets a /64's (128-64=>64 bits of address space). Larger entities get /48's (128-48=>80 bits of address space). So everyone (you, your device, your network) has plenty of addresses to spare.
Addresses are written in hex, in the form 0011:2233:4455:6677:8899:aabb:ccdd:eeff (that's 8 groups of 16 bits -> 128 bits). If there are zeroes in an address on, you can omit them (ie. 1122:3344:5566:0000:0000:0000:0000:1234 => 1122:3344:5566::1234).
Some protocols (ARP, kinda DHCP, ICMP) are merged into a single procotol called NDP (neighbor discovery protocol).
When you're a home user, your router/gateway/CPE would get a /64 from your ISP, for example 2001:1234:4567:89ab::0/64. to use for your local devices. For example, in your LAN, your router would probably be 2001:1234:4567:89ab::1, while your devices would probably be addressed between 2001:1234:4567:89ab::2 and 2001:1234:4567:89ab:ffff:ffff:ffff:ffff.
If your router is not using DHCPv6, but instead only NDP, it would simply respond with a “we're using 2001:1234:4567:89ab::0/64 here, I'm your gateway” to every device coming into the LAN, and the device would choose an address based on it's NIC MAC and some other facts. If your router would be using DHCPv6, the exchange would be just like in ordinary DHCP (dhcp request -> dhcp reply with IPv6 address to use).
There is some extra (standarized) link-local magic with IPv6 - if you seen an fe80::0/16 address on your link, then it means that there was no IPv6 router/DHCP server on there and the devices in the LAN are establishing connection independently - but of course, without any access to the rest of the IPv6 Internet. This is also done via NDP, IIRC.
And that's basically it. No NAT. The devices in your LAN are directly routable to devices in your friend's LAN, or to Facebook's server
Do I want to turn it on - do I get some benefit, or would I just do it for the greater good? I've heard it's actually better for the spying agencies?
Some of the nicer profits of IPv6 for hacker is that you can host servers at home, and expose them to friends/clients.
As for "spying", yes someone (a website you connect to) can figure out whether it's you or your roommate who are connecting. But they can figure that out by browser headers, anyway. This is also partially mitigated by RFC4941.
Is there anything in all this that can make my machine behind the router get the same address all the time? Do I manually assign it locally as with IPv4 and hope no one else in the building assigns the same IP?
Is there anything to assign IP-to-Hostname at least on the local (behind the gateway) level. Or is this handled via some other service I need to run (maybe some server that runs dns and allows machines to push their desired hostname to it).
one caveat: DHCP6 is younger than IPv6, so not all router firmwares will have gotten around to implementing it yet.
On the machine I'm writing this reply on, when I boot it up (with either FreeBSD, Linux or Windows) it gets a router advertisement from the router (which is routing both IPv4 and IPv6). It gets a standard NATed DHCP lease for IPv4. For IPv6, it combines the /64 network prefix assigned by my ISP with the NIC MAC address to compute a unique (and stable) address.
There are of course alternatives. I can assign it statically, I can use privacy extensions to periodically change the address (seems to be the default on Windows for me) or I can use DHCPv6 to use DHCP-allocated addresses (my router doesn't support this though I have done it in the past with BIND).
I'd argue that many, if not all of the issues with v6 back in 2002 have been resolved, and that v6 adoption is accelerating, both on eyeball networks as well as on public-facing internet services.