RE: decisions on L2 solutions documents

Hi Hamid,  thanks for the comments.  Please see in-line.  regards, Neil

Hamid Ould-Brahim wrote 02 June 2003 23:00
Neil, 
> > 
> > Wouldn't it be kind of cool if we could use the same 
> > addressing format for 
> > what ever MPLS LSP/pseudo wire/tunnel? Is it too simple to 
> > say we always 
> > identify the end-point of an LSP/PW/tunnel with a unique 
> > address (I mean 
> > unique for the LSP/PW/tunnel so no 2 PWs share the same address)? 
> > 
> But it that case how about pw created for l2vpns. 
> NH=> Hamid why do you think you need PWs at all?  


Actually I don't need them :-).
NH=> Yes I know that too ;-)

I see them as a tool 
for l2vpns (among others)...and Peter mentioned addresses 
for pw...so I asked :-)...
NH=> You are right....this was an oversight on Peter's case.  He/I and many
others here are of the same view as the one I expressed.

>The only 
> reasons PWs exist 
> is as follows: 
> -     to fix the problems caused by mp2p LSPs.....if you want any-any 
> connectivity use a cnls mode, in a co pkt-sw mode only p2p and p2mp 
> constructions make any arch sense....mp2p leads to problems 
> and kludges; 
> -     to keep the pretence-up that MPLS =~ IP.....if true why 
> do we need 
> MPLS then?  Use the right mode for the right application....both are 
> required; 
> -     to squeeze a bit of the server-layer BW via client 
> compression (a 
> total non-issue IMO when other technolgies squander BW). 
> 
> Would these 
> addresses be required to be always unique or can be 
> overlapping?  
> NH=> PWs are *not* a layer network....they are actually a set of 
> encapsulation functions that belongs in the client->server adaptation 
> process (network-interworking case). 
Isn't control/signaling, establishment of pws part of pwe3 work 
besides just encapsulation...
NH=> Yes.  But think about TCP (or any other 2-point session-like protocols)
these need 'control' but they are not 'networked entities'.  The mere reason
this is being done does not justify PWs existence.  Their existence is
justified in other less technical ways....my prior comments indicate what I
mean by this.


> Client->server 
> adaptation should be a 
> minimalist mapping with (ideally) no functions that need 
> management (eg 
> sequence number aberrations).  PWs don't need addresses, ie 
> PWs are not 
> switched network entities like LSPs...its MPLS that needs access point 
> addresses in its own right since MPLS creates a layer 
> network.  Further, 
> when one considers mixed partition service-interworking 
> between technologies 
> belonging to the *same* network mode (eg 
> FR_access<=>MPLS_core<=>ATM_access) 
> PWs simply disappear anyway.  I know service-interworking is 
> not on IETF's 
> (PWE3) agenda, but speaking as an operator its very high on 
> ours to drive 
> down capex/opex so we can have a common core per network 
> mode.....so please 
> (as a vendor) don't ignore it. 
> 
Not at all ignoring it. I am actually in favor of addressing 
interworking case and I don't see all l2vpn problems need to be solved 
only through pwe3 to be taken as a complete package. In fact BGP 
signaling is an example that pwe3 encaps can be useful without pw 
signaling scheme (that's maybe a different topic)...
NH=> I think so.  I think BGP is an excellent protocol for distributing
reachability information (all modes), esp in a client/server case for a VPN,
but I don't think its a good general signalling protocol.....this has to be
mode specific.
 
> Its the above issues that sit behind Peter's original comments. 
> 
Okay. Not sure I understand the client-server terminology 
and mpls/pwe3 layer subtleties...I guess this topic is more appropriate with

PWE3 layer discussions and MPLS layer debate going on 
in pwe3 wg. 
NH=> Agreed....but they are kind-of interwined as I am sure you know.  Have
a word with Dave Allen in your organisation on these topics....he has an
excellent grasp of the issues and they are very important to us anyway.