The MPLS WG Archive

[Kireeti Kompella <kireeti@juniper.net>]

Hi Eduard, Juha,

> This is indeed a bit confusing. 

Yes, sorry, my fault.  When the text says "CE configuration", what
is meant is "per-CE configuration on a PE" ...

> As far as I understand (Kireeti: correct me if I'm wrong here) the
> parameters are actually *configured* on the PE.

Correct.

> However, the CE should know
> about this.

How much the CE needs to know can vary from as little as "this interface
(to the PE) is Frame Relay, and the PE will tell me which DLCIs are up,
and inarp will tell me addresses/OSPF will tell me routes" to "for this
interface, this is the list of DLCIs and corresponding IP addresses" if
LMI/inarp is not supported.

> In particular the CE configuration should reflect the same logic
> that is applied by the PE in mapping CE-to-CE circuits to eachother. This is
> the algorithm described in Section 4.3.1. When CE and PE configuration are
> not aligned on this point, the CE may expect to connect to one site on a
> particular circuit, while the PE actually connects it to another one.

If the CEs are running (say) OSPF, and learn routes dynamically,
they should not really care whether DLCI 56 or 93 gets them to a given
site.  However, the network operator may want to know this.

> I think the text should say that the parameters are configured on the PE,
> and that the CE should know about this / reflect this in it's configuration.
> Currently, it is described in the exact opposite way. Also it makes less
> sense to configure it on the CE since there is no interaction between PE and
> CE to exchange this information.

The next version will state this more clearly.

Kireeti.