Hi Eric,

> -----Original Message-----
> From: Eric C Rosen [mailto:[email protected]]
> Sent: Wednesday, July 29, 2015 12:35 AM
> To: Xuxiaohu; [email protected]; [email protected]; BESS
> Cc: [email protected]; [email protected];
> [email protected]
> Subject: Re: [Idr] [bess] 2 week adoption call for
> draft-rosen-idr-tunnel-enaps-00.txt (7/6 to 7/20/2015)
> 
> On 7/24/2015 4:40 AM, Xuxiaohu wrote:
> > Now it seems that the above rationale has been shaken by this draft.
> >
> The argument in favor of the Encapsulation-SAFI doesn't seem to have been as
> persuasive to everyone else as it was to the authors of RFC 5512 ;-)
> 
> > Futhermore, since it allows to attach BGP Encap attributes to routes
> > now, the BGP encapsulation extended community can now be safely
> > replaced by BGP Encap attribute. Therefore, I wonder whether RFC5512
> > would be obsoleted by this draft in the end.
> 
> Quite a few people have suggested that RFC5512 should be obsoleted by the
> tunnel encaps draft; this is worth considering.
> 
> However, I think we need to maintain compatibility with the parts of
> RFC5512 that have actually been implemented, such as the encapsulation EC.
> 
> > Since BGP encapsulation attributes are now attached to routes, the
> > AFI/SAFI of the NLRI is already capable of indicating the payload type
> > of the advertised tunnel protocol.
> 
> I don't think the payload type can necessarily be inferred from the AFI/SAFI 
> of
> the UPDATE that is carrying the Tunnel Encapsulation attribute.  Consider, for
> instance, a case where the next hop is not of the same address family as the 
> NLRI,
> but the route to the next hop is the route carrying the Tunnel Encapsulation
> attribute.

It seems that the example that you had mentioned is the usage of the Tunnel 
Encapsulation attribute e as described in RFC5512, rather than the usage of the 
Tunnel Encapsulation attribute as proposed by this draft:)

> > it seems unnecessary to allocate two additional type codes for
> > MPLS-in-UDP and IP-in-UDP.
> 
> I think I could argue this point either way:
> 
> - On the one hand ... When you are about to sending a particular payload 
> packet
> through a tunnel, you know whether the packet is MPLS or IP, so you know
> enough to form the encapsulation correctly.
> 
> - On the other hand ... if you want the control plane to form the rewrite 
> string in
> advance, you might want to form a different rewrite string for MPLS-in-GRE
> than for IP-in-GRE.

If the Tunnel Encapsulation attribute associated with a labeled unicast route 
indicates UDP tunnels, should it be obvious that MPLS-in-UDP is supported by 
the originator of that labeled unicast route? Similarly, If the Tunnel 
Encapsulation attribute associated with a unlabeled IPvx unicast route 
indicates UDP tunnels, should it be obvious that IPvx-in-UDP is supported by 
the originator of that unlabeled unicast route?

Best regards,
Xiaohu

> I would be good if we had some general rules for when it is good to define a 
> new
> tunnel type and when it is good to reuse an existing tunnel type.  Of course, 
> we
> still need to maintain compatibility with the tunnel types that are already
> defined.

_______________________________________________
BESS mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/bess

Reply via email to