4 ms·

Yes migration to IPv6 is essential, but virtual hosting for websites and NAT for ISP subscribers provides a way to indefinitely postpone IPmageddon. This never

Yes migration to IPv6 is essential, but virtual hosting for websites and NAT for ISP subscribers provides a way to indefinitely postpone IPmageddon. This never seems to be mentioned. Also once a block is assigned the actual usage is unpredictable and not reported on that I'm aware of - I know several folks who have blocks assigned by their DC and they're only using a fraction.

AndrewDucker 12y ago
So far as I can tell NAT for ISPs is brought up pretty much every time that IPv6 is mentioned on HN. And every time people point out that it breaks the fundamentally decentralised nature of the internet and makes it harder to run P2P services. And also, pretty much every time someone mentions unused blocks, and someone else points out that even if we reclaimed the largest known blocks that aren't publicly routed this would only expand the timeline for a few months.
- tormeh 12y ago
  >it breaks the fundamentally decentralised nature of the internet and makes it harder to run P2P services Well, those are two things ISPs don't care about...
  - krakensden 12y ago
    Yup. Comcast's ideal world is where they sell TV via IP for $200/month + Amazon.com access with a referral fee, and block everything else.
    - agwa 12y ago
      Comcast has laudably been on the forefront of deploying IPv6. And several other major consumer ISPs in the U.S. are working on deploying it as well (and in the interim, AT&T is offering tunneling with 6RD). Amazon, on the other hand, still does not support IPv6. I browse the Web from Comcast with a Firefox extension that tells me if a site is using IPv6 or not, and the lack of websites supporting IPv6 concerns me way more than the pace of IPv6 rollout by consumer ISPs. There's simply no conspiracy here.
      - marcosdumay 12y ago
        There isn't much reason for servers to support IPv6 yet. Clients can use 6to4, and will use it as soon as they become IPv6 only, because... well, there isn't much reason for servers to support IPv6 yet. (Yep, that's circular.) Servers will add support for IPv6 in due time, when admins get used to having it on their machines, and all network equipment supports it, and it stops breaking the sites for people on IPv4 only and people behind slower IPv6 tunnels.
        pdkl95 12y ago
        > Clients can use 6to4 And with that, the transition of the internet from a peer to peer network to a client/server model is finally completed. /not directed at you specifically; I'm just depressed from how efficiently the empower, democratic, anybody-can-publish qualities of the internet are being undone
        marcosdumay 12y ago
        We've already completed that transition when ISPs started blocking ports from their customers. Putting everybody in IPv6 will make things better, and odds are that it will create an entire new crop of descentralized services. But current http and smtp servers gain very little from supporting IPv6 just now.
        tacticus 12y ago
        Actually there is a nice reason for any reasonable sized stack to support v6. it makes life easier. you run globally addressable ranges that just make life easier as you get to remove all those wonderful dnats and snats in your networks. as your internal ranges become easier to manage and more plentiful so you can layout your networks in ways that make sense. group services in IPs that make sense without ever really worrying about exhausting those ranges. comcast for example moved to v6 because they ran out of rfc1918 to run their administrative networks in.
        p1mrx 12y ago
        I think you mean NAT64/DS-Lite/MAP (which are all types of carrier-grade NAT), since 6to4 is just a means of connecting two IPv6 endpoints across an IPv4 tunnel. This photo sums up the problem with relying on CGN: https://www.flickr.com/photos/jfesler/6751925977/ https://www.flickr.com/photos/jfesler/6751925977/ When an ISP funnels multiple users through the same IPv4 address, you can no longer use IP-based bans without inflicting collateral damage. One way around this problem is for both the ISP and the website to support bigger addresses.
  - ay 12y ago
    Running a box that smoothly maintains hundreds of millions of sessions (a household might would peak at about 1000 concurrent sessions) is a different expense than running boxes that just statelessly route the packets for those hundreds of millions of sessions.
    - p1mrx 12y ago
      You actually don't need state to share an IPv4 address, if you statically allocate port ranges to customers, and route packets based on the address+port. But rather than build special routers that understand port ranges, it makes more sense to map each address+port into an IPv6 prefix, and use standard IPv6 routing. That's the basis of MAP: http://en.wikipedia.org/wiki/Mapping_of_Address_and_Port http://en.wikipedia.org/wiki/Mapping_of_Address_and_Port Of course, the scheme is useless if you don't have an IPv6 network in the first place.
      - ay 12y ago
        Oh, indeed! The basic argument is "keeping state is more expensive and fragile than not keeping it". BTW, speaking of MAP: the CPE-side code is now in OpenWRT trunk. I have built a few images here, in case someone is interested to test: http://stdio.be/blog/2014-06-10-OpenWRT-MAP-images-from-pure-trunk/ http://stdio.be/blog/2014-06-10-OpenWRT-MAP-images-from-pure... (Needless to say, would love to hear feedback)
muppetman 12y ago
It annoys me you're being downvoted. I don't necessarily agree with you either, but I see nothing worthy of a downvote here.

AndrewDucker 12y ago

So far as I can tell NAT for ISPs is brought up pretty much every time that IPv6 is mentioned on HN. And every time people point out that it breaks the fundamentally decentralised nature of the internet and makes it harder to run P2P services. And also, pretty much every time someone mentions unused blocks, and someone else points out that even if we reclaimed the largest known blocks that aren't publicly routed this would only expand the timeline for a few months.

tormeh 12y ago
>it breaks the fundamentally decentralised nature of the internet and makes it harder to run P2P services Well, those are two things ISPs don't care about...
- krakensden 12y ago
  Yup. Comcast's ideal world is where they sell TV via IP for $200/month + Amazon.com access with a referral fee, and block everything else.
  - agwa 12y ago
    Comcast has laudably been on the forefront of deploying IPv6. And several other major consumer ISPs in the U.S. are working on deploying it as well (and in the interim, AT&T is offering tunneling with 6RD). Amazon, on the other hand, still does not support IPv6. I browse the Web from Comcast with a Firefox extension that tells me if a site is using IPv6 or not, and the lack of websites supporting IPv6 concerns me way more than the pace of IPv6 rollout by consumer ISPs. There's simply no conspiracy here.
    - marcosdumay 12y ago
      There isn't much reason for servers to support IPv6 yet. Clients can use 6to4, and will use it as soon as they become IPv6 only, because... well, there isn't much reason for servers to support IPv6 yet. (Yep, that's circular.) Servers will add support for IPv6 in due time, when admins get used to having it on their machines, and all network equipment supports it, and it stops breaking the sites for people on IPv4 only and people behind slower IPv6 tunnels.
      - pdkl95 12y ago
        > Clients can use 6to4 And with that, the transition of the internet from a peer to peer network to a client/server model is finally completed. /not directed at you specifically; I'm just depressed from how efficiently the empower, democratic, anybody-can-publish qualities of the internet are being undone
        marcosdumay 12y ago
        We've already completed that transition when ISPs started blocking ports from their customers. Putting everybody in IPv6 will make things better, and odds are that it will create an entire new crop of descentralized services. But current http and smtp servers gain very little from supporting IPv6 just now.
      - tacticus 12y ago
        Actually there is a nice reason for any reasonable sized stack to support v6. it makes life easier. you run globally addressable ranges that just make life easier as you get to remove all those wonderful dnats and snats in your networks. as your internal ranges become easier to manage and more plentiful so you can layout your networks in ways that make sense. group services in IPs that make sense without ever really worrying about exhausting those ranges. comcast for example moved to v6 because they ran out of rfc1918 to run their administrative networks in.
      - p1mrx 12y ago
        I think you mean NAT64/DS-Lite/MAP (which are all types of carrier-grade NAT), since 6to4 is just a means of connecting two IPv6 endpoints across an IPv4 tunnel. This photo sums up the problem with relying on CGN: https://www.flickr.com/photos/jfesler/6751925977/ https://www.flickr.com/photos/jfesler/6751925977/ When an ISP funnels multiple users through the same IPv4 address, you can no longer use IP-based bans without inflicting collateral damage. One way around this problem is for both the ISP and the website to support bigger addresses.
- ay 12y ago
  Running a box that smoothly maintains hundreds of millions of sessions (a household might would peak at about 1000 concurrent sessions) is a different expense than running boxes that just statelessly route the packets for those hundreds of millions of sessions.
  - p1mrx 12y ago
    You actually don't need state to share an IPv4 address, if you statically allocate port ranges to customers, and route packets based on the address+port. But rather than build special routers that understand port ranges, it makes more sense to map each address+port into an IPv6 prefix, and use standard IPv6 routing. That's the basis of MAP: http://en.wikipedia.org/wiki/Mapping_of_Address_and_Port http://en.wikipedia.org/wiki/Mapping_of_Address_and_Port Of course, the scheme is useless if you don't have an IPv6 network in the first place.
    - ay 12y ago
      Oh, indeed! The basic argument is "keeping state is more expensive and fragile than not keeping it". BTW, speaking of MAP: the CPE-side code is now in OpenWRT trunk. I have built a few images here, in case someone is interested to test: http://stdio.be/blog/2014-06-10-OpenWRT-MAP-images-from-pure-trunk/ http://stdio.be/blog/2014-06-10-OpenWRT-MAP-images-from-pure... (Needless to say, would love to hear feedback)

muppetman 12y ago

It annoys me you're being downvoted. I don't necessarily agree with you either, but I see nothing worthy of a downvote here.