The MPLS WG Archive

[Bernard Zelmans <bzelmans@calynet.com>]

let me describe what I'm thinking of and let me know if it's stupid.
basically the idea is to distribute the initiation of the LSP for the VC (as
per Martini definition), closer to the customer. so I could have a simple L2
device that would be MPLS enable so that it learns from the PE (through LDP)
the FEC information. Then you would map all ethernet traffic coming on one
port or one VLAN id onto that LSP. On the PE you would receive this labelled
packet and would tunnel it to the peer egress point. so basically you don't
change the protocol (Martini or Kompella) but you extend to another hop the
initialisation of the LSP (corresponding to the VC or inner label once in
the tunnel from the PE)and advertise the FEC through LDP from the PE to this
MPLS enable device.
the point is that I could implement PE function on my box but which one to
go to (Martini or Kompella)? it looks to me that Kompella has more
advantages. but it also looks like that there will be 2 different
implementations out there.
Thanks in advance for your input,

b

-----Original Message-----
From: Sasha Vainshtein [mailto:Sasha@AXERRA.com]
Sent: Wednesday, June 27, 2001 10:34 AM
To: Bernard Zelmans; mpls@UU.NET
Cc: Alik Shimelmits; Gonen Zilber; Israel Sasson
Subject: RE: L2 MPLS based VPN


Bernard,
Inner labels in both Martiti-Transport and Kompella L2 VPN drafts 
are distributed directly between peer edge devices, and have meaning
only between these devices. Both Martini and Kompella drafts allow using
targeted LDP sessions for this purpose, and Kompella also allows using
MP-BGP for this purpose.
For the aggregation device to allocate the outer label, it has to be able
to infer somehow the FEC of the incoming packet.
I see the following options for the packet transmitted by the access device
to the aggregation one:
1. It is a label packet with the outer label in the stack being the inner
label
of the virtual connection. This label cannot be valid in the aggregation
device
(as it did not participate in its distribution) so it either drops the
packet or, even
worse, misuses it.
2. It is an IP packet (say, with MPLS-in-IP protocol type and 
the inner label following the IP header). In this case the access device:
* Has to operate as an IP router
* Has to know IP address of its remote peer device
* The aggregation device may push the "outer" label on top of IP header
or just to forward the unlabeled
*The remote access device must recognize the incoming IP packet as 
one carying MPLS-in-IP, strip its IP header and then process the incoming
label.
I.e., the 2nd option will work, but it imposes restrictions on access
devices. E.g.,
you cannot connect Ethernet switches (which are not IP routers) in this way.
PE devices in both Martini and Kompella are assumed to be LERs (as defined
in RFC 3031)
with additional capabilities allowing support of encapsulation of L2
packets; this approach
does not impose any restrictions on the access (presumably, CPE) devices.
With best regards,
                                   Sasha Vainshtein
email:     sasha@axerra.com <mailto:sasha@axerra.com> 
tel:       +972-3-7659993 (office)
           +972-8-9254948 (res.)
           +972-58-674833 (cell.)
 


> -----Original Message-----
> From: Bernard Zelmans [mailto:bzelmans@calynet.com]
> Sent: Wednesday, June 27, 2001 4:49 AM
> To: mpls@UU.NET
> Subject: L2 MPLS based VPN
> 
> 
> Hi all,
> 
> looking at both Martini and Kompella draft on L2 based VPN, 
> the LSP with the
> inner+outer label is started from the same node (PE). Is 
> there a particular
> reason for that? why don't we allow an access device to 
> initiate the inner
> label LSP and have an aggregation device start the tunnel?
> Thanks in advance,
> 
> b
>