Re: [mpls] ELI and EL in P2MP (Re: [PWE3] ELI as a reserved label)

[Curtis Villamizar <[e-mail]>]

In message <[e-mail]>
David Allan I writes:
>  
> Thanks Curtis, that clarifies it for me...we are still only spreading
> over a P2P span...
>  
> Cheers
> D 


Yes.  If you consider a ECMP LSP as per draft-kompella-mpls-rsvp-ecmp
to be a P2P span, which it certainly is in the "client layer" sense
(to borrow from ITU-speak, but apply it to MPLS without the TP).

I agree with you (and Shane, and even John Drake :) that applying this
to P2MP more generally where it does not re-merge is not viable.

Curtis


> -----Original Message-----
> From: [e-mail] [mailto:[e-mail]] 
> Sent: Wednesday, August 04, 2010 10:08 AM
> To: David Allan I
> Cc: Yong Lucy; 'Shane Amante'; [e-mail]; [e-mail]; [e-mail]
> Subject: ELI and EL in P2MP (Re: [mpls] [PWE3] ELI as a reserved label)
>  
>  
> In message <[e-mail]>
> David Allan I writes:
> >  
> > A clarification please...
> >  
> > I can understand the utility of an entropy label on a P2MP PW in a 
> > MS-PW case, as I can load spread the underlying unicast segments 
> > between the S-PEs...is that the practical use case?
> >  
> > Loadspreading a p2mp LSP such that it becomes an p2mp2mp seems rather 
> > fraught with peril...
> >  
> > In which case an entropy label only really applies to a p2p label..
> >  
> > Am I missing something?
> > D
>  
>  
> Dave,
>  
> A flow label is per draft-ietf-pwe3-fat-pw is applicable only to PW.
>  
> An entropy label (EL) is applicable to LSP as well as PW.  The entropy label indicator (ELI) just marks where the EL is in the stack to remove any ambiguity.  This is per draft-kompella-mpls-entropy-label.
>  
> A P2MP LSP could carry many traffic flows, such as many video distribution streams, or many IP multicast groups (<s,G> groups as P2MP is defined).  If so, an entropy label could be added, though a hash on s,G would have the same effect.
>  
> If a P2MP LSP carrying IP multicast crosses a LAG today, it is split based on s,G since s,G is the same as unicast source and destination, with dst being in the range of multicast addresses.  Since that is the case, P2MP LSP and ELI/EL could be applicable to link bundle (with all ones component) and RSVP-TE ECMP (draft-kompella-mpls-rsvp-ecmp), both of which branch but merge and deliver all traffic to the same node.
>  
> Curtis