Re: [Idr] Comment on draft-zhang-idr-upstream-assigned-label-solution

[[e-mail]]

Hi Eric,


Eric C Rosen <[e-mail]> 写于 2015/07/30 23:24:24:

> On 7/30/2015 5:42 AM, [e-mail] wrote:
> 
> >      Refer to RFC5331, the context-specific label can be auto 
generated or
> > manually configured. In IPv6 environment, manually control is 
required.
> 
> Are you referring to Section 8 of RFC 5331, "Context Label on LANs"? 
> The labels discussed there are 20-bit values that uniquely identify a 
> router within the context of a particular LAN.  These are really a form 
> of domain-wide label, with the "domain" being a single LAN.  As I said 
> earlier, your draft really seems aimed at the problem of dynamically 
> assigning domain-wide unique labels, so I would agree that assigning 
> context labels on LANs is a possible application.
> 
> Note that the material from Section 8 of RFC 5331 is only applicable if 
> you are using ethernet multicast frames to carry MPLS multicast packets. 

>   I'm not sure whether there are actually any deployments of that.
> 
> > And the network service that is supposed by L3VPN, not only include 
MVPN,
> > but also include EVPN, DC interconnection, and so on.
> 
> The LAN context label only identifies a router, it is not 
> service-specific.  So you only need one per node per LAN.    (I think 
> it's a lot like a "node SID" in SPRING or a "BFIR-id" in BIER.)
> 
---I don't think that the context label is a lot like SID or BFR-id. The 
context 
label can be used to distinguish the service, such like MVPN, EVPN, DC 
interconnection, VxLAN and so on. So only one context label identifies a 
router
is not enough. More context labels are needed for different services on a 
router.

> The issue of how best to assign an identifier that is unique within a 
> certain scope is an issue that comes up in many contexts, though usually 

> it is IP addresses rather than 20-bit identifiers that need to be 
> assigned.  There are a lot of prior solutions that could be considered, 
> generally not requiring either BGP or time synchronization.
> 
> > And the p2mp and mp2mp tunnel may be used to transfer extranet
> > flow. When
> > different multicast flows which are forwarded to receivers in 
different VPN
> > are mixed into one P-tunnel, are there more conflict?
> 
> MVPN extranet does not impose any particular requirement for 
> upstream-assigned labels or domain-wide labels.

---The different flows share one P-tunnel, that will decrease the number 
of P-tunnels.
And it will make more easier for service and policy deployment of 
operators. Maybe MVPN
extranet should consider the use of upstream-assigned labels. 

Sandy