[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: Stateless Prefix Delegation I-D updated (draft-savolainen-stateless-pd)
Hi,
On Tue, March 2, 2010 16:56, teemu.savolainen@nokia.com wrote:
> Rather the delegating router would delegate a prefix minus one /64. I.e.
> the one /64 would not be really assigned back to the originating link, but
> rather never assigned to requesting router at the first place...
In theory what you propose should work, but...
Let's assume we assign 2001:db8:1:200::/56 to the device.
It would give the uplink 2001:db8:1:200::/64 and say 2001:db8:1:201::/64
to its WLAN port. For the device itself everything is fine - its routing
table contains two orthogonal /64 routes.
However the Access Concentrator has to deal with a paradoxical situation:
several overlapping routes. If we assume the AC has host-ID ::1 and the
CPE device or mobile phone has ::2:
2001:db8:1:200::1/128 loops back to the AC(*)
2001:db8:1:200::/64 has the link as direct target without gateway
2001:db8:1:200::/56 goes to gateway fe80::2 via the same link
(*)some IPv6 stacks may omit this one
It may work or it may not work. The alternative is a lot more entries in
the routing table.
What I described below has the upside that the third route does not
overlap with the others. It has of course the downside that it uses up
slightly more address space.
>> The solution is to assign a separate portion of the provider prefix to
>> direct link prefixes and delegate something different from this. It
>> requires a bit more thinking on the network operators part, but causes
>> much less pain for the support hotline... ;-)
>>
>> Eg. if the provider uses 2001:db8::/32 and has about 10mio customers
>> (24bit customer ID) one solution would be to reserve 2001:db8:0::/40
>> for
>> link assignment and 2001:db8:100::/40 through 2001:db8:ff00::/40 for
>> PD.
>> As a customer I could for example get 2001:db8:11:2233::/64 via SLAAC
>> for
>> the uplink and 2001:db8:1122:3300::/56 via DHCPv6 PD for delegation
>> downstream.
>
> Isn't this close to what I describe in stateless draft, i.e. having a /32
> bit ISP prefix, and then using 24 lowest bits of the /64 prefix given to
> UE (with SLAAC) in calculating the delegated /56 prefix? In DHCPv6-"light"
> this calculation would be done by DR, in Stateless this would be done by
> RR.
Correct.
Konrad