Re: new proposal for rpslng

[Florent Parent <]

--On 2002-01-09 16:45:39 -0500 jhaas@localhost wrote:

> >    peering-set: AS1-v6
> >    peering: AS1 afi ipv6 3ffe:ffff::1 at afi ipv6 3ffe:ffff::2
> We should, if possible, restrict this such that peering with different
> afi's on router-expression-1 and router-expression-2 is a syntax error.
That certainly makes sense. noted.

> >    typedef:      # Subsequent address family values
> >                  # Currently defined: unicast and multicast
> >                  safi_elm
> >                  enum[unicast, multicast]
> you're missing unicast-multicast.  SAFI 3 is distinct from safi 1 and 2
> despite the fact that it should be functionally identical.  But a
> valid filter of "import safi unicast" and "import safi multicast"
> could ignore "import safi unicast-multicast".
Correct. I will add this safi so that this follows the same safi's as 
defined in rfc2858. (as a side note: If we want to support mpls-vpn (or 
other) RPSL policies in the future, we could just add the appropriate safi 
here)

> It should also be noted that most implementations allow an action
> of installing a route learned from one safi into another as well.
> This allows routes learned from a unicast source to be installed
> as only multicast routes or both unicast and multicast.
Do you see that as injecting routes from one safi to another ? If so, it 
could be handled with something like the following :

  import:  [protocol <protocol> [afi(address-family)
                                 safi(subsequent-address-family)]]
           [into protocol <protocol>[afi(address-family)
                                 safi(subsequent-address-family)]]

There would need to be some sanity checks here, like making sure the same 
AFIs are used.  Let me know your thoughts and examples on this.

Thanks for your feedback Jeff.

Florent.