Hi, all
I recently read the draft
draft-ietf-pce-disco-proto-igp again, here are some comments。
When IGP is used as a PCED protocol, the
capabilities of pce will be advertised periodicly along with other lsa/lsp
information. In my opinion, unless the PCE's capabilities change, there
is no need for the PCC to receive and handle the information
again.
JLLR: But these are standard IGP procedures. This is
the same for link states, they are refreshed even if there is not change. By
the way the refresh interval can be quite large (e.g. 18 hours for
IS-IS).
Furthermore, if most of the routers in a
domain will always have no request to a PCE, there is also no need for them to
maintain the information.
JLLR: In this case PCE discovery is deactivated on
these routers and they will not process the PCED and PCES
TLVs...
My concern to PCE-DEST-DOMAINS sub-TLV is that we
can't anticipate how many destination domains a PCE can computate path
towards, as a result,the length of space taken by this capality in the
message can't be controlled.
JLLR: Actually you have the same problem with the
number of TE-links advertised by an LSR or the number of reacheable
prefixes advertised by an ABR.
One
may limit by configuration the number of dest domains
advertised.
So I think
if the PCC send a request to a PCE which can't compute the path for the reason
of the destination domain, the PCE can tell the PCC in the PCrep
message the right PCE.
JLLR: But how the PCE will have the
information?
One
additional quesion: is it suitable for the PCE to have such ability to know
exactly the destination domain it can compute towards?
JLLR: This is actually required. For instance in
an inter-AS case, an AS may be connected to multiple ASs, and there may
be multiple PCEs in this AS (e.g. ASBRs), each responsible for computing path
towards a given neighbord AS.
Some functions specified in the PATH-CAMP-CAP sub-TLV
are also defined in the PCEP, a clear division should be
made.
JLLR: Actually, at
the time being the PCEP spec does not define any capability
information.
In
current PCEP spec, we only mention that detailed capabilites could be carried
in optional sub-TLVs of the Open object (carried in the Open
message).
But
you are right such capability information may be carried in
PCEP.
The
idea would be to carry simple capabilities in the IGP
and more complex capabilities in PCEP.
Anyway I don't really see any issue if we define two ways to carry some pieces
of information:
-IGP-based capability discovery is useful for PCCs
that don't have a PCEP session with the PCE.
-PCEP-based capability discovery is useful when IGP based discovery is
not activated.
Thanks for these
comments,
Regards,
JL
Cheers,
Zhang
Renhai
