Re: Extending a /64

[[e-mail]]

Tony,

I intended for this thread to explore if there was interest in exploring other solutions than longer prefixes than /64.
Although also including that one.
Just doing longer prefixes doesn't solve the "permissionless extension" problem either.

Btw, nothing stop you from not supporting SLAAC. You can make your addressing plan any which way you like.

Best regards,
Ole

> On 8 Nov 2020, at 14:00, Tony Whyman <[e-mail]> wrote:
> 
> The problem of the /64 limit came up in ICAO working groups earlier this year when developing a global addressing plan for the ATN/IPS and there would be strong support here for allowing subnet prefixes to go beyond the /64 boundary.
> 
> Our problem is that we wanted to define an addressing plan that uses 39 bits to identify each aircraft, allows for a minimum of 4 more bits for subnetting and works within the existing standards. I'll give some background below for those interested in why 39 bits, but the problem is that if you are going to allocate a /64 (deriving from a /60 MNP) to each subnet on an aircraft then you need a /21 as your initial allocation - and this is before you start thinking about other ATN/IPS users such as drones.
> 
> In order to avoid a potential problem getting a sufficient address allocation, we did look at extending into the "other 64 bits" of an IPv6 address for subnet IDs, which was looking very "unused". After all, you can only cram in a handful of systems into the avionics bay of even the biggest aircraft and, if you extend the addressing scheme into the back cabin, passengers can only bring on so many mobile phones, tablets and laptops in their hand luggage. 64 bits is an overkill for the host identifier and, if only we could have allocated (e.g.) a /96 for each aircraft subnet, then the problem would just go away.
> 
> In the end, we concluded that we could not bend the existing standards to do this and did not have the desire to push for change. Hence, we are in the process of asking IANA for a /16 for the global civil aviation community. Hopefully we will get this. However, an argument is expected given that the current policies push back against such a large allocation and demand a utilisation efficiency that we will never achieve. However, if we don't get a /16 and the standards stay as they are, then a private address space and NAT at the boundaries may be the only answer - not really what anyone wants.
> 
> I would certainly support raising the existing limit. A hard limit of a /96 would seem to be a good idea to stop ISPs going too far. It's probably also worth noting that perhaps one day every home on the planet may require an IPv6 Prefix. That is perhaps over a billion given the world's population. With densely packed address allocation that still requires 30 or so bits. Once you take into account any kind of geographical sub-allocation then you will need a lot more. A /96 would seem to be much easier to live with than a /64.
> 
> Tony Whyman, MWA
> 
> Background
> -----------------
> 
> For operational reasons, the ATN/IPS address allocation should allow for a common prefix for all aircraft operated by the same airline. Network diagnostics and firewall rules are often cited in support of such a requirement. It is also highly desirable that there should be a common prefix for all airline IPv6 Address Prefixes. Again this is to support simple firewall rules. Resisting DoS attacks is extremely important in our environment and these include firewall rules that prevent packets from external sources being sent to aircraft unless authorised, if only to minimise the risk of overloading limited capacity wireless subnetworks. If every airline and ATC Centre has a different address prefix then managing these rules will be an almost impossible task if they all have unrelated prefixes. A common ICAO prefix is thus highly desirable as are common prefixes for each category of address space user.
> 
> It is believed that 15 bits is the minimum necessary to sub-allocate a prefix to each airline registered with ICAO. This leverages an existing registration scheme and allows for a reasonable degree of growth.
> 
> There are also very good reasons for using the existing ICAO 24-bit aircraft identifier as part of an aircraft''s /60 MNP i.e. to sub-allocate each airline's address space to each of their aircraft. For very good safety reasons, ATC Centre Flight Data Processors need to correlate datalink messages (callsigns and 24-bit), surveillance reports (radar and ADS-B - which also use the 24-bit scheme), voice  messages (using callsigns) and Flight Plans. Basically, you don't want to introduce yet another identifier into the mix and the use of overlapping schemes improves confidence in the overall result.
> 
> There is also the small matter of backwards compatibility with the existing ATN/OSI CLNP addressing scheme. This also uses an airline identifier/24-bit approach to address allocation and if we keep to the same address semantics then it becomes feasible to introduce the ATN/IPS with minimum impact on the high certification environment of the Flight Data Processor. Any alternative will cost a lot more and put back deployment for several years.
> 
> So, we really are boxed in and 39 bits + 4 bits for aircraft subnets is the minimum needed for assignment of a /64 to each aircraft subnet. Hence, the need for at least a /21, which ends up as a request for a /16 when you add in other users and a desire for nibble boundaries.
> 
> On 08/11/2020 10:25, [e-mail] wrote:
>> Starting a new thread.
>> 
>> A problem described in variable-slaac is:
>> 
>> "It should be possible to extend an end-user network that is only assigned a /64"
>> 
>> I believe that is a problem worth looking at.
>> This problem is not only restricted to the mobile access case, think connecting a host with VMs to a link.
>> 
>> The address delegation to a site problem is intertwined with the autonomous networking problem of the site itself. The IETF solution is DHCPv6 PD + HNCP. The expectation of addressing of a network is that the addresses are long-lived.
>> 
>> There are many potentional solutions:
>> 
>> a1) ask the network operator for more address space.
>> a2) change provider
>> a3) introduce government regulation
>> b1) steal the uplink /64 (64share)
>> b2) steal multiple /64s from uplink
>> c) overlay. use e.g. LISP to tunnel across the access ISP to connect to an ISP that support multi-homing and larger address space.
>> d) MultiLink Subnet Routing. I.e. let a single /64 span multiple links. draft-thubert-6man-ipv6-over-wireless, draft-ietf-ipv6-multilink-subnets
>> e) NAT
>> f) P2P Ethernet. Hosts are not on the same physical link, so let's stop pretending they are. A consequence of that is that links don't need subnets. Only assign addresses to hosts. draft-troan-6man-p2p-ethernet-00
>> g) extend the /64 bit boundary. HNCP implementations do /80s I think (forces DHCP for address assignment)
>> 
>> 
>> Requirements:
>> R-1: Permissionless. Not require an action on the network operator
>> R-2: Arbitrary topology
>> R-3: Long-lived address assignments
>> R-4: Support bad operational practice: flash renumbering / ephemeral addressing
>> 
>> 
>> Is there interest to work on this problem?
>> If so, suggestions for next steps?
>> 
>> Best regards,
>> Ole (without any particular hat on)
>> --------------------------------------------------------------------
>> IETF IPv6 working group mailing list
>> [e-mail]
>> Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
>> --------------------------------------------------------------------
>> 
> 
> --------------------------------------------------------------------
> IETF IPv6 working group mailing list
> [e-mail]
> Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
> --------------------------------------------------------------------