The MPLS-OPS Archive

[Spice Sylvia <falsesylvia@yahoo.co.uk>]

inline...

 --- "Morris, Stephen" <Stephen.Morris@marconi.com>
wrote: 
> Hi,
> Comments below.
> Best wishes
> Stephen
> 
> -----Original Message-----
> From: mpls-ops-request@mplsrc.com
> [mailto:mpls-ops-request@mplsrc.com]On
> Behalf Of Spice Sylvia
> Sent: 14 February 2005 11:45
> To: mpls-ops@mplsrc.com
> Subject: [PM-SPAM] RE: [PM-SPAM] Re: [MPLS-OPS]:
> GMPLS based VPNs
> 
> 
> Hi,
> 
> 
>  --- "Morris, Stephen" <Stephen.Morris@marconi.com>
> wrote: 
> > Hi Sylvia
> > Local significance avoids the need for allocating
> > globally significant
> > labels. This reduces the end-to-end burden of path
> > reservation, let's say
> > I'm an LER and I want to use label x  for an LSP.
> If
> > this was a globally
> > significant label then all affected nodes would
> have
> > to agree on this label.
> 
> 
> simply let the label represent the "end point", how
> is
> it different from vanilla IP routing or GRE/IPSEC
> tunnels then?
> 
> SM> Label as endpoint is no longer MPLS! The label
> then has to grow with the
> address space - scalability problem
> 

Split it across 2 routers, use filters/communities in
BGP or RSVP TE or  directed LDP.
if you want to go from "a" to "b" each node in the
middle has to know how to get to "b", is that correct
or not?
So your label may simply be tag to the next-hop, but
each node in the middle has to know the general
direction of the next-hop, yes or no?
So if you setup and LSP from A to B, each node along
the over which u signal the LSP from A has to know how
to get to B. right?
What you can do is break the set so that " B is known
only by some nodes in the path, the path being
contiguous" and the path to B will obviously be from
that set of nodes. correct?

SO whatever you do, you cannot eliminate "the lookup"
at some point in the network (the LER in this case) as
long as you "look at something which tells you what
the destination is" at the LER, correct?

Now think how your telephone network works, there is
no "destination marker " in the "data path". It is all
in the control path, correct? what you do loose is the
"connectionless/udp kind facility".


> > This would impose an unnecessary cross-network
> > burden.
> > 
> 
> what burden? How do we provision enterprise networks
> in the absence of MPLS? either P2P ipsec tunnels, or
> leased lines and cisco makes more $$ with 3600  and
> 2600 routers :) or we use VRs capped on FR/ATM
> cores.
> 
> SM> If a global label represents a path, then all
> nodes in the path must be
> configured to   recognise that label. This is an
> extra burden on path
> computation and signalling, e.g., if LER A uses
> label x then I assume your
> scheme requires that x be used on all nodes in the
> path. All nodes must
> agree to use this global label x. This constraint is
> an extra burden on the
> nodes

If the end point is identifier by "X" then all nodes
in the path have to either know
a. a mapping between X and a label (which means they
still have to know X)correct?
b. you could hide X on some nodes using label
stacking, and end up with MTU and SAR issues. correct?

> 
> > Re: why label switching: if I understand your
> > question, label switching
> > separates control and forwarding, so the control
> can
> > be made arbitrarily
> > complex, while at the same time leveraging legacy
> > proven protocols. 
> 
> which "legacy proven protocols??" are we talking
> about
> here?
> 
> SM> The layer 3 protocols.

There is no layer, think of a wire between A and B, A
C and A D, simple wires with the same vertex, how does
data flow?
you have to look at something at A to identify what
the path must be.
right?
similarly, each node in the way has to know each end
point even for IP/ or a subset of all nodes have to
have contiguous path to the end points.
The end points still have to be unique..correct?

> 
> >This
> > moves the network away from complex legacy
> > technology and in the direction
> > of a richer service-oriented model.
> 
> 
> i personally like the "parallel universe concept" of
> virtual routers. You may want to look at :
>
http://www.space1999.net/~chronicle/ExE/Year1/atap.html
> and
>
http://www.amazon.com/gp/product/product-description/0679746048/104-9024539-
> 4159138?_encoding=UTF8&n=283155
> 
of course the above is bull crap :)) but its just the
difference between "multihop and tunnels" in a way ;)


	
	
		
___________________________________________________________ 
ALL-NEW Yahoo! Messenger - all new features - even more fun! http://uk.messenger.yahoo.com

-------
The MPLS-OPS Mailing List
Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
Archive: http://www.mplsrc.com/mpls-ops_archive.shtml