- Terminology: You define "Node" and "Message Publisher" but the
  difference is not very clear as they are both defined to be
  publishers. My interpretation is that your "Node" encompasses one or
  more "Message Publishers", i.e., it is e.g., a device. Note that a
  paragraph down we have 'publisher node' and 'network node'. (Do we
  really need a definition for 'node' if it is what I think it is?)

- I expected that the section "Motivation" expands on the motivation
  for this work but instead it starts straight with lost and corrupt
  YANG notification messages and data integrity and then motivates
  just the "Message Publisher ID". Despite the rather specific
  content, I am also surprised how the introduction of an ID helps
  with data integrity. Back to the document structure, we do you
  discuss the need for a "Message Publisher ID" before the solution
  overview? Could it just be a part of the solution overview?

- Why does a Receiver have to reassemble packets? (And are we talking
  about packets or messages here?) Perhaps s/packets are reassembled
  and decapsulated/notification messages are decapsulated/?

- Section 6: How does the Message Publisher ID ensure message
  integrity?

- Figure 1 could perhaps be extended to show that also the master is
  sending notifications related to the subscriptions to the receiver.

- How are global subscriptions disassembled into component
  subscriptions? I guess this is somewhat implementation specific,
  perhaps say so, i.e., make it clear what is covered by the standard
  and what is specifically not. I have no idea whether there are
  scenarios where interoperability between a master publisher and a
  publishing agent not originating from the same vendor or project
  is required.

- A client subscribing to notifications may not be aware whether the
  implementation is distributed or not, correct? May I assume that
  care has been taken that an unaware client will be fine when it hits
  a distributed publisher?

- Are there any restrictions on the uint32 number space for publisher
  IDs? Is 0 a valid publisher ID? Would it perhaps make sense to
  reserve 0 for situations where it can be useful to express 'no
  publisher ID'? Would it make sense to introduce a common typedef for
  the message ID that is then used everywhere rather than a plain
  uint32?

- I wonder whether the module name 'ietf-distributed-notif' is good.

- I have not checked or even validated the XML snippets in the
  appendix.