The MPLS WG Archive

[Juha Heinanen <jh@lohi.eng.telia.fi>]

eduard metz writes:

 > however, there
 > are many situations in which this is not the case. for example: when
 > the network is treated as a collection of point-to-point links (with
 > individual /30s), or when the network is native l2 ...
 > in these cases some alignment / coordination between pe and ce
 > configuration is required. you need to know which circuit is
 > connected to which site to end up with a correct, and working
 > configuration.

this too can and has been automated at least in case of atm. you simply
have one-to-one mapping between subinterface numbers and vpi numbers and
then when the ce learns a new label, it knows which subinterface it
belongs to based on its vpi component.

the same can be done with native mpls by using label stacking: the outer
label selects the vpn and the inner label determines a site within that
vpn.  it doesn't matter if a "vnp" has many sites or if it is just a
point to point link.

-- juha