[As an individual and co-author]
In the section 5.3, "PCE Discovery", you said "Mechanisms
that rely on advertising or querying PCE locations across
domains or provider boundaries are undesirable.".
I want to know why it is an "undesirable" way.
For the record, we has in the WG, long discussion on whether
we should advertise PCE across domains using for example BGP
for that matter.
There was a good consensus not to, considering the limited
number of interconnected domains. The advantage was limited
and introduces other issues (security) ...
Right. I recall that discussion.
And the situation with hierarchical PCE is worse.
Consider, a child PCE in domain-A does not know where the Parent PCE is. It
could be in *any* domain. In order to make contact with the parent PCE, the
child must either know where the parent is (configuration as recommended in
the I-D), or advertise itself across *all* domains. That is, for
child-advertisement to work, every PCE in the network must broadcast an
advertisement to every other node in every other domain in the whole
network. This is "undesirable".
Alternatively, consider a parent PCE that wants to contact the PCEs in each
of domain-A, B, ..., Z. It could be configured (as recommended in this I-D)
or it must send a broadcast query message to every node in each domain.
Since there is no domain address aggregation(CIDR being just a dream) the
parent PCE must broadcast to every node in the entire network. This is
"undesirable".
That is, both approaches to discovery are simply not scalable, and would
probably be filtered out at domain boundaries anyway.
With hierarchical PCE, we are looking at a peering relationship that is
certainly administrative and probably commercial. Such relationships involve
a lot of configuration and negotiation. Adding the configuration of
parent/child PCE locations is an almost trivial addition.
Cheers,
Adrian
_______________________________________________
Pce mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/pce
