(addressing Peter here, mostly, but expanding on AJ’s points)

On 25 January 2015 at 10:28:19, Alastair Johnson (aj@sneep.net(mailto:aj@sneep.net)) wrote:
> On 1/24/15 11:11 AM, Peter Lambrechtsen wrote:
> > My personal view is I personally doubt there will be much movement for
> > the next 5+ years or even more unless the "killer v6" app comes out or
> > something major happens like Facebook only supporting v6 from a certain
> > date. There are fairly simple reasons why.
>
> I find that a depressing view. There's no reason for most ISPs why IPv6
> should not be delivered on broadband today, especially from new market
> entrants. 

The killer app is reducing cost of operating an IPv4 network, as AJ covers and as I talk about some below.

> > Most ISPs have plenty of spare address space and the retail fixed
> > broadband market is pretty static.
> > Even the new players like bigpipe and myrepublic have managed to get v4
> > address space. And those two didn't launch with a dual stack. Why was
> > that???
> > It's been shown that even the new players can survive with CGNat in the
> > days of Section 92 without needing to log everything I presume. 

There are various solutions for this that are in use by operators I’ve worked with, the main being relying on allocating blocks of ports, rather than a port per connection etc. This makes logging easier, so that S.92 obligations can be met. You have to be careful to choose a block size that balances logging capacity with public address:customer ratios, but that’s not too hard.

Having said that, logging all connections is actually not that difficult if you’re not afraid to cut a bit of code, just don’t do what someone will inevitably suggest and stick it all in to Splunk :-)

> > In the mobile space I can see it happening. But at the same time CGNat
> > seems to be working for them too.
>
> CGN will not save you, and will add significant capital and operational
> cost to any ISP, regardless of technology chosen.
>
> IPv6 has an up-front cost to complete the engineering to support it; CGN
> has costs forever. I do not think that is a wise long term investment -
> and as Brian pointed out, the world's largest operators seem to have
> agreed with that view.

This is the basis of how I help people justify IPv6 in their org. IPv6 puts a cap, albeit an unknown cap, on CGN costs. Without IPv6 the CGN cost/demand/etc. will grow with Internet usage. With IPv6 the CGN cost will grow with IPv4 usage only, and so in theory will stop growing at some point as IPv6 content becomes a larger % of total traffic.
Additionally, as large amounts of traffic for NZ providers comes off CDNs that support IPv6, that CDN traffic never has to be NATed (Akamai, Google both support IPv6, if you’ve got them in your network I’m sure you already know this). This is also a driver to encourage customer adoption, of course - the more customers pushing v6 packets, the less NAT you have to do.

This is I suppose not such an issue if you feel you have enough IPv4 addresses, and so never need to NAT. Given all the mobile carriers (I’m excluding MVNOs here, as I don’t know) in NZ NAT their customers by default, we can infer that having enough addresses simply isn’t true.

I would be surprised if these sort of considerations are not on Spark’s radar, given CGN can be quite costly and complicated at scale.

It would be interesting to get stats on number of handsets that support IPv6 on your network. These numbers wouldn’t be hard to get, and you should with a bit more work be able to figure out how much traffic these devices do to IPv6 capable CGNs, and then translate the lack of running IPv6 in to $ value based on CGN capacity. Do that every month and start plotting a lines, along with cost to roll out IPv6. The sooner you do it, the better, but at some point the roll out will pay for itself in CGN costs so quickly that you can easily put together a business case.

--
Nathan Ward