Re: Well off the original thread topic... draft-bonica-tunneltrace-02

[Giles Heron <giles@packetexchange.net>]

David Allan wrote:

> Hi Giles...long time no hear...;-)


yeah, I've been resisting the temptation join the melee.  Don't always 
succeed of course... ;-)

 
> Another thread I have my regrets about....comments in line...
> 
> <snip>
>  >
>  > Actually there are other things you can do with "just" LDP that you
>  > can't do with IP.
>  >
>  > 1)  draft-martini etc.  Sure, IP equivalents of this are now being
>  > defined (UTI^H^H^H L2TPv3 et al) but they aren't as widely available,
>  > and are much less efficient encapsulation-wise (as they require an IP
>  > header plus a tunnel header instead of a couple of labels).
> 
> If you're discussing:
> - i) using extended LDP adjacencies to hand out PW labels
> - ii) the relative effy. of MPLS labels vs. IP headers
> no argument, but that is somewhat orthogonal to the discussion of the 
> value that LDP-DU tied to routing may or may not provide.


In fact it is the use of LDP-DU (or RSVP-TE) to hand out the tunnel 
label that gives us efficiency wrt L2TPv3 et al (and also seems to give 
us the availability of code - few of the vendors support L2TPv3 or 
draft-martini over GRE at this point - though of course practical issues 
like availability of code should be kept out of IETF discussions...)


>  > 2)  MPLS VPN (RFC2547).  I won't go into the pros and cons of
>  > this right
>  > now ;-)
> 
> Current implementation is MPLS but I don't think there is a sustainable 
> dependency there...LDP permits the undelying MPLS network to behave like 
> an IP network. Seems to me they should be interchangable.


they are - the above issues notwithstanding.


>  > 3)  BGP free network core.  This may improve stability
>  > (though there's
>  > certainly a strong argument that BGP is less destabilising than LDP
>  > right now) and security (the core doesn't need to know how to
>  > get to the
>  > Internet.)  Also this allows you to layer multiple distinct
>  > IP backbones
>  > over one MPLS core.
> 
> Left that one out of my analysis, not sure how simple LDP helps here. I 
> thought you LSP meshed your BGP speakers (which would be explicit 
> routes) and eliminated IGP.


Yes, you LSP mesh the BGP speakers over LDP-DU tunnels using the IGP 
that is carried through the core.  So the core doesn't need to know 
about the BGP routes.


>  > > It is when I get into topology manipulation independent of dynamic IP
>  > > routing that MPLS starts to add value and to claim this is not
>  > > connection oriented use of MPLS forwarding IMHO smacks of the same
>  > > "tunnel"/"layer" semantic debate that can also be said to
>  > waste much time.
>  >
>  > this is your view.  And it is certainly one view.  But it isn't
>  > necessarily the only valid view ;-)
> 
> The "topology manipulation" part or the "terminology part" ? ;-)


the "starts to add value" part - I think :)

Giles


> later
> Dave
>  
>  
> 


-- 
=================================================================
Giles Heron    Principal Network Architect    PacketExchange Ltd.
ph: +44 7880 506185              "if you build it they will yawn"
=================================================================