[6bone] separating IPv6 experimental from production traffic

Nicolas DEFFAYET nicolas.deffayet@ndsoftware.net
22 Aug 2002 00:06:55 +0200


On Wed, 2002-08-21 at 22:52, Rik van Riel wrote:
> On Wed, 21 Aug 2002, Ronald van der Pol wrote:
> > On Wed, Aug 21, 2002 at 10:29:15 -0700, Bill Manning wrote:
> >
> > > 	Tell me how you propose to do this?
> >
> > I think we first have to decide what to do: cleanup or separation.
> 
> This shouldn't be too hard, actually.  The main (only) complaint
> I've seen is bad routing over the experimental network, to be
> more precise, bad routing of _production_ packets over the 6bone.
> 
> Here is my naive proposal:
> 
> 1) everybody can still have ipv6-over-ipv4 tunnels like today
> 
> 2) peers notify each other whether their network is production
>    or experimental
> 
> 3) experimental (3ffe::/16) pTLAs announce only 3ffe:: prefixes
>    to their production (2001::/16) peers
> 
> 4) experimental pTLAs announce all prefixes to their 3ffe::/16
>    peers
> 
> 5) production networks announce all prefixes to everybody

I add:
  
  6) Don't provide by default full transit to peers, provide full
transit only if peer request it.


It will be very difficult to apply a policy on 6bone without a big
clean.

> 
> This means that packets from one production network to another
> production network will only get routed over production networks
> and NOT over potentially unstable experimental networks.
> 
> It also means that the production and experimental parts of the
> ipv6 space still have full connectivity to each other, ie the
> reachability of either 3ffe::/16 or 2001::/16 space hasn't
> gotten any worse.
> 
> The only change is that traffic between production networks is
> travelling through production networks only and doesn't depend
> on possible failures in the experimental network.
> 
> This also gives an incentive for ISPs to migrate from experimental
> to production prefixes, since that will potentially give them better
> ipv6 connectivity.
> 
> Any flaw in my reasoning ?
> 

I think that it's the best way.

Best Regards,

Nicolas DEFFAYET