Hi Kexin,

Thanks for the comments, please see below.


/Jan


On 11/1/11 11:48 PM, "[email protected]" <[email protected]> wrote:


>I think the objective of the two draft are the same that stateful PCE
>synchronized with the state of LSPs although by different mechanism.

The objectives of draft-crabbe-pce-stateful-pce are twofold:

* Keep a stateful PCE synchronized with the state of LSPs in a PCC; this
objective is the same as the objective of draft-tang-pce-stateful-pce.
* Allow a stateful PCE to determine the timing of LSP setup in the network
(in addition to determining the LSP paths through the network).

We call the PCE that only wishes to know the LSP state the 'passive
stateful PCE'. We call the PCE that controls the timing of LSP setups in
the network the 'active stateful PCE'.

>
>You defined two new PCEP messages (PCRpt and PCUpd), while we extended
>the existing PCNtf message.

We considered extending the PCNtf message, but in the end decided to to
define a new message for LSP State Reporting (PCRpt). We defined the PCUpd
message to satisfy the other objective of draft-crabbe-pce-stateful-pce ­
the active control of LSPs by a PCE.

There are good reasons to define a new message for LSP State Reporting.  I
think we are both agreement that an LSP state report/notification must
carry a set of LSP attributes that describe the LSP's state. Now, if we
want to re-use the PCNtf message for LSP state reports / notifications, we
can either define in the NOTIFICATION Object a set of optional TLVs that
can carry the LSP attributes, or redefine the PCNtf message itself to use
already defined PCEP Objects to carry LSP attributes.

The TLV approach is not ideal: we already have a set of PCEP Objects that
can carry the LSP attributes, now we would have to define all of them as
TLVs as well. And if new LSP attributes are introduced to the PCEP
protocol in the future, the may have to be defined both in PCEP Objects
and in NOTIFICATION TLVs.

The approach where the PCNtf message is redefined is not ideal either. For
one, we may be breaking existing implementations because we are redefining
the message and not using optional TLVs to define new functionality.
(Note that with a new message we are not breaking existing
implementations, because new messages are only used if stateful
capabilities are negotiated between PCEP Speakers). Second, the PCNtf
message as currently defined in PCEP seems to be designed for PCEP events
(PCE congestion, cancellation of Pending Requests, Š), and changing its
semantics to carry both PCEP events and LSP state does not seem to be the
right thing to do. Finally, redefining the PCNtf message changes its
structure to a point where we may as well define a new message, and let
each message perform just one function well.


>

_______________________________________________
Pce mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/pce

Reply via email to