IPv6 PI allocation

Niels Bakker niels=cluenet at bakker.net
Thu May 17 19:52:40 CEST 2007


>>>> RFC3178-style multihoming will create more entries in the global 
>>>> routing table once a network using that connects to an IXP.
>>>not at all.  RFC3178 does not require ISPs to announce any specific 
>>>routes (other than /29 or /32 of the ISP).
>> RFC3178 exists in some kind of fantasy world.  Quoting from it, "3. 
>> Basic mechanism": "We assume that our site is connected to 2 ISPs, ISP-A 
>> and ISP-B."
>> 
>> This is an unrealistic requirement for the reasons I spelled out (the 
>> need to connect to an IXP,
>no need to connect to any of IXes (by IXP you mean Internet eXchange 
>Points - am I right?).

No need?  No possibility, you mean.  The need exists on the current 
Internet (in both technical and business climate), why do you think that 
would disappear?

(What else could I mean with IXP in the context of multihoming?)


>> plus the need for independence from any ISP's 
>> network
>this is not fullfilled by RFC3178, that's true.  see below.

>> - you cannot withdraw an announcement from a particular broken 
>> upstream except by pulling all DNS records).
>you can workaround this with RFC3178.  the assumption is that ISP-A 

No you cannot.  If ISP-B disappears (and with that I mean their 
/29-to-/32 disappears from the global routing table) half the session 
setups to hosts within the RFC3178-multihomed site will time out.


>and ISP-B are having transit relationship with each other, that's all.

That would rule out end sites connecting directly to two tier-1s.

I remain convinced that RFC3178 describes a technically sound solution 
that is acceptable to a very small minority of sites looking to 
multihome.


	-- Niels.


More information about the ipv6-ops mailing list