Hello,

please find below my OPS-DIR review.
The draft is well written and easy to read but would benefit from the explanation of some decisions taken in the design and statements provided without a justification. I really like the extensive usage of examples, which clearly explain situations that might be hard to be understood using the text only.

The synchronization among multiple stateful PCEs and avoidance of computation loops is for sure a valid use case but i was wondering why the hierarchical PCE approach is not taken into account (it's also explicitly said in the document). Wouldn't it be easier to have a hierarchical PCE coordinating the various PCEs instead of having them talking to each others? I guess the authors have done this analysis already and providing a comparison between the two approaches in the draft would be beneficial. 

Some more comments related to specific parts of the text below:

- Abstract: For completeness, a stateful PCE can be used also for GMPLS, not only for MPLS-TE.
- Intro: "he hierarchical PCE use case is out of the scope of this document." see my general comment above.
- Intro, figure 1: "when there is a change in LSP1, the PCC should report to PCE1.  From PCE2's perspective, PCC1 reporting the update of LSP1 to PCE2 is lower than sync it from PCE1 to PCE2." This statement should be justified. Why the communication PCC1-PCE2 should be slower than PCE1-PCE2 ?
- Section 3.1 only includes 3.1.1 and no extra text, probably you don't need 3.1.1 and just 3.1 is fine.
- Section 3.3: " When propagating LSP state changes from a PCE to other PCEs, it MUST ensure that a PCE always uses the freshest state coming from the PCC." i'm not sure i understand who the subject is, and how it can ensure that the freshest state is used.
- Figure 6: Lenght isn't 16? How long is the LSP state DB version number? 64 bit? do you expect that many version number updates during the lifetime of an LSP?
- Section 3.5: You say taht PCC doesn't need to be aware of teh computation priority. Then who tells the PCE what is the priority of each association group?
- Section 6: The need of a Path-Vector TLV makes me think that the hierarchical approach would be simpler. With the hierarchical approach you wouldn't need a message loop detection mechanisms.

Thanks
Daniele