APNIC IPv6 transit exchange

[Iljitsch van Beijnum]

On 29 nov 2007, at 18:08, Jeroen Massar wrote:

> Well, I did wrote the level above it and do agree with the part below,
> the below part is from Michael Horn, thus please fixup your quoter ;)

I always assume people pay attention so I don't need to repeat  
everything. The extra >>'s should be enough.  (-:

> Next to that they are also suggesting people do *TRANSIT* over the  
> link.

Ugh. Do we know the AS number they'll be using for the MLPA router yet  
so that we can filter out all paths with that AS in it?

> As such, any
> organisations willing to offer FULL TRANSIT will be greatly  
> appreciated."

This is the traditional RIPE peering policy which is nice for a RIR  
but not for an MLPA router.

>> Some kind of multipoint tunnel, such as 6to4, would be more
>> appropriate here.

> It definitely would, except that there is one problem with 6to4 which
> makes it useless for this case, as one can't do:

> ip ro add 2001:db8::/32 via 2002:c000:022a::1

Why wouldn't you be able to do that?

If you simply set up BGP peering using 2002:... addresses, BGP will  
view 2002::/16 as one big IPv6 subnet so all other 2002:... addresses  
are directly connected and you can get the appropriate MLPA behavior,  
i.e., if 2002::1 is the MLPA router and 2002::2 and 2002::3 only peer  
with 2002::1 but not with each other, traffic from 2002::2 to 2002::3  
will see a next hop of 2002::3 even though it learned the prefix from  
2002::1.

There is of course the slight complication that 6to4 is also used for  
other stuff which you may not want to expose your BGP routers to...

> Yes, point-to-point tunnels between every site with BGP on top of  
> that.

Too configuration-intensive...