News for hackers. Stuff that matters.
Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by ArchOversight on Monday March 06, 2023 @04:47PM (#35048132)
The ISP said they got only a limited set of IPv4 addresses, those are assigned to their CGNAT gateways. Which are expensive, especially for an Indian Reservation, not the deep pockets of some MSP.

The users are behind CG-NAT.

But instead of using IPv6 which is cheaper (no need to maintain CG-NAT, translation devices, or deal with traffic that is being routed way more expensively) the Roku devices are only streaming over IPv4.

Each new user that adds an IPv4 only device adds additional load the CG-NAT and additional capacity will need to be provisioned. That is an additional expense and burden.

Most of my traffic at my house (on Comcast) is over IPv6, because most if not all streaming services now support IPv6 for content delivery, so the small amount of data that may need to go over IPv4 when the majority can go over IPv6 reduces the load on IPv4.

Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by brokencode on Monday March 06, 2023 @05:02PM (#35048334)
Why does the IPv4 address need to be publicly exposed outside of a customer’s LAN? I thought you could set up NAT on a router to translate local IPv4 addresses to something IPv6 that is exposed publicly.

Is this simply bad planning from the ISP where they didn’t handle it correctly? Or is there something I’m not understanding about NAT?

I think in an ideal world all devices would be using IPv6. But I thought it would be common knowledge among network engineers that many devices still use IPv4, so you have to either handle it somehow or tell your customers that some of their devices simply won’t work.

Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by ArchOversight on Monday March 06, 2023 @05:12PM (#35048478)
Because the IP address the Roku is trying to reach is an IPv4 address. You can't just translate that to IPv6 and say "good luck little packet".

If you translate at the customers router that's fine and all, but now you have an IPv4 packet in an IPv6 packet, that IPv6 packet needs to get routed to a device that knows how to then turn it back into an IPv4 packet so that it can then go travel on the open internet like the electrons intended...

Once that IPv4 response come back, it needs to get translated back to IPv6, sent to the customers edge, which translates it back from IPv6 to IPv4 to send to the Roku device.

Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by brokencode on Monday March 06, 2023 @05:22PM (#35048598)
Oh, what a mess. I don’t like that at all.

I was assuming that there was some way to translate the IPv4 address of the server to an IPv6 one and process it that way, putting the burden of supporting IPv6 on the server side. I had no idea that Roku would actually need to be exposing an IPv4 server to handle these requests.

That makes sense then that the ISP would need some number of IPv4 addresses that it could use to communicate with IPv4 servers then on behalf of IPv4 client devices.

Shame on Roku for perpetuating this problem.

Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by makomk on Monday March 06, 2023 @06:30PM (#35049364)
The more usual reason why devices would be making an IPv4 request despite having an IPv6-supporting connection is precisely because the server doesn't support IPv6 - either the network it's on doesn't support it, or it's not configured for it, or more rarely the software on the other end doesn't have IPv6 support. The request can't just be translated into an IPv6 one because there's no way of knowing that the server even understands that, let alone what its associated IPv6 address is.
Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by zamadatix on Monday March 06, 2023 @05:04PM (#35048376)
CG-NAT (NATing centrally at the ISP's internet edge) is cheaper/simpler than something like 464xlat (NATing v4 locally over v6) since you can do the former on 2 boxes instead of 20,000. That said the 2nd option is much cooler :).
Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by ArchOversight on Monday March 06, 2023 @05:10PM (#35048443)
464XLAT still requires IPv4 boxes on the edge to translate IPv6 traffic back to IPv4. Whether that is two boxes or 20,000, the same is true for other CG-NAT solutions. Someone somewhere is bearing the cost of translating.
Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by zamadatix on Monday March 06, 2023 @05:17PM (#35048526)
Usually you need a special box at the consumer side to do 464XLAT, it's not something you can just ask your customer's Netgear to do and it's usually more expensive if you want to provide it as the customer's rented router. CG-NAT however looks completely normal to all gear (other than the particular numbers assigned) except the 2 edge boxes. It's the ultimate cost saving kludge.
Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by cmeacham98 on Monday March 06, 2023 @05:22PM (#35048592)
Most modern OSes (I know for a fact Android, iOS, and Windows) support automatically doing the 4->6 translation on their side (as a matter of fact, some cellular networks in the US are ipv6 only).

I'm unsure if consumer routers would pass on the appropriate RA flag to tell the OS they need to do this in their default configuration however.

Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by zamadatix on Monday March 06, 2023 @05:29PM (#35048668)
Well the devices that support 464xlat already support IPv6 by definition. CG-NAT/464XLAT+Client NAT at the gateway is for the devices like the Rokus that don't support IPv6 at all.
Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by cmeacham98 on Monday March 06, 2023 @05:30PM (#35048684)
NAT64 doesn't work if the connection is made to a hardcoded IP(v4) address rather than DNS entry, and that's way more common than you'd think! Thus why 464XLAT and similar exist.
Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by zamadatix on Monday March 06, 2023 @06:03PM (#35049074)
Absolutely but CG-NAT would be NAT444 (i.e. double NAT to dual stack) not NAT64. I mean they probably did try NAT64 based on the description of the botched 1st attempt but CG-NAT would have negated the need for NAT64 in the first place. Putting a x464 CLAT that can gateway/NAT the local network everywhere would definitely achieve the same functional goal though, just more ISP devices.
Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by zamadatix on Monday March 06, 2023 @05:00PM (#35048311)
Realistically they are already running a large scale NAT device anyways, otherwise the v6 only clients wouldn't be able to reach v4 only internet services (hello HN server), and if they had planned to CG-NAT from the start it probably could have all been done on the same pair of internet edge routers for much less than 300k additional expense.
Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by Symbiote on Monday March 06, 2023 @05:12PM (#35048473)
As the article says, the Roku traffic is the overwhelming majority of that ipv4 traffic.

HN and similar mostly-text websites would barely show up on the statistics.

Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by zamadatix on Monday March 06, 2023 @06:59PM (#35049661)
NAT scales in both dimensions, you're just as likely to need a larger box for a larger connection table as you are a larger bandwidth.
Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by noAnswer on Monday March 06, 2023 @07:30PM (#35049905)
At some point you also need more IPv4-addresses. There are only 65535 ports to be mapped.
Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by zamadatix on Monday March 06, 2023 @07:54PM (#35050066)
At some point but remember it's not "after 65535 connections per IP" it's "after 65535 connections of the same type to the same IP per IP". That is to saythese NATs where 156.78.92.154 is a single public address on the NATing box:

  156.78.92.154:6781<->8.8.8.8:53 UDP
  156.78.92.154:6781<->8.8.8.8:53 TCP
  156.78.92.154:6781<->8.8.4.4:53 TCP
  156.78.92.154:6781<->8.8.4.4:53 UDP
Can all exist simultaneously even though only 1 port is in use and we haven't even started changing the destination IP or ports yet (e.g. 80 and 443 as the destination won't double count and thousands of web servers on different addresses can be accessed via the same source port).

With CG-NAT you can usually support somewhere around 30k customers on the free /24 you get from ARIN for having only IPv6 and needing space to translate in (for new orgs like new ISPs this is assigned immediately out of a reserved block, bypassing the waitlist for existing orgs trying to get free reclaimed space).

Re:Roku devices don't support IPv6 in 2023 and its costing ISPs by ArchOversight on Tuesday March 07, 2023 @10:01AM (#35056067)
It's also about ports. If I have 10 Gbps of IPv4 flow but my devices are licensed for and operate on 1 Gbps, I need to license 10 1 Gbps ports.

When I was last working at an ISP, various CG-NAT solutions charged not just based on the connection table, but also there was various licensing for various slide-in cards.

If the ISP is having to buy additional hardware/additional ports on upstream providers just to power their CG-NAT that is an additional cost.

If the POP where you have your CG-NAT doesn't have more bandwidth available you end up running your ports hot... so now you either need to rent a new location, get your connectivity up there and start figuring out how to route the traffic.

It's not as simple, and it all costs copious amounts of money. Especially if your IPv6 can more easily be distributed across multiple POP's with multiple forms of connectivity and traffic shaped using BGP or other solutions thereby reducing the load on a single upstream port.