Hi Benjamin,

Inline:

> On 17 Jul 2019, at 19:18, Benjamin Kaduk <[email protected]> wrote:
>> 
>> ----------------------------------------------------------------------
>> DISCUSS:
>> ----------------------------------------------------------------------
>> 
>> The "Peer Up message Information" TLV type seems under-specified and
>> under-motivated.  (It is not mentioned in Abstract or Introduction.)  Why
>> does it need to be defined in this document, and what role is it
>> expected to play?  Who is the expected audience for it?  Is it limited
>> to the "group name"-like functionality described in Section 7.1?  Why is
>> cleartext appropriate, and are there any potential privacy
>> considerations for any potential use cases?
>> 
>> I’d +1 the reply from Jeff on this.
> 
> While Jeff's reply does help me personally to understand the motivation,
> I'm left wondering what role I expect the final RFC itself to play in this
> matter.  I think I'd like to see a brief mention in the Introduction that
> this TLV is added in order to convey administrator-defined information
> about the rib-out; is that consistent with your stance on the matter?

Hoping it is not felt i am abusing this reason, i’d echo my comment on the new 
stats types: my feeling is that mentioning this optional TLV in the 
introduction is too much detail and while this is integral part of the draft, 
it is not a core concept. Shall, then, you ask me then what is the core concept 
of the draft, very legitimate question at this point, i’d tell you: adding new 
vantage points from which monitor BGP data. And abstract and Introduction both 
describe this concept at different - as it would be expexcted - resolution of 
detail. My proposal - also because to reach the point of describing the TLV one 
would have to speak about filtering and multi Loc-Rib - is to leave mention / 
full elaboration of this TLV in the relevant sections (6.3.1, 7.1 and 9.3) 
hoping they convey the same reasoning that Jeff mentioned in his previous email.

>> ----------------------------------------------------------------------
>> COMMENT:
>> ----------------------------------------------------------------------
>> 
>> Section 1
>> 
>>                      An example of pre-policy verses post-policy is
>>  when an inbound policy applies attribute modification or filters.
>>  Pre-policy would contain information prior to the inbound policy
>>  changes or filters of data.  Post policy would convey the changed
>>  data or would not contain the filtered data.
>> 
>> Can applying policy ever act as injecting new data?
>> 
>> Yes. Jeff Haas proposed the edit below to draft-ietf-grow-bmp-loc-rib-04 — 
>> would the same/a similar edit address your point?
>> 
>> [ .. ]
>> 
> 
> I'm not 100% sure  I'm looking at the right link -- the highlighted lines
> are about requiring the Table Name to be sent when multiple filters are
> applicable.  My question was intended to be about the wording here, about
> whether "post policy would convey" should mention the possibility of new,
> injected, data (in addition to changed or removed data).

Apologies, having done multiple edits, the lines changed. I was referring to 
the following piece of text:

“In case of any change that results in the alteration of behaviour of
an existing BMP session, ie. changes to filtering and table names,
the session MUST be bounced with a Peer DOWN/Peer UP sequence.”

Although this would be more relevant to the case of a policy change - i realize 
now, after your last email, a tangent.

To your point, i feel we are good, let me explain. Slightly before in the 
introduction we have:

"The Adj-RIB-In post-policy conveys to a BMP receiver all RIB data
after policy filters and/or modifications have been applied.”

Which says _all_ RIB data (that is: new, changed or removed). Then, continuing, 
an example is given:

“An example of pre-policy versus post-policy is when an inbound policy
applies attribute modification or filters."

Related to this we have the concluding sentence:

“Post policy would convey the changed data or would not contain the filtered 
data.” 

Where “changed” does not refer to new/changed/removed RIB data but rather to 
modified by the policy.

Having done some homeworks i feel i can also say: this paragraph was pretty 
much copy/pasted by rfc7854 :-)

>> [ .. ]
> 
> [ .. ]
> 
>>  o  Timestamp: The time when the encapsulated routes were advertised
>>     (one may also think of this as the time when they were installed
>>     in the Adj-RIB-Out), expressed in seconds and microseconds since
>>     midnight (zero hour), January 1, 1970 (UTC).  If zero, the time is
>>     unavailable.  Precision of the timestamp is implementation-
>>     dependent.
>> 
>> Are leap seconds included in this value?  (Yes, I see that this language
>> is basically taken directly from RFC 7854, which also left it
>> underspecified.)
>> 
>> I would propose to leave it underspecified as well, leaving it 
>> implementation specific.
> 
> I can understand that decision, even if I'm not fully comfortable actively
> endorsing it.

I see, i am open to suggestions on this point.

>> [ .. ]
> 
> [ .. ]
> 
>> Section 9.3
>> 
>> The registry where this seems to be listed claims to be the "BMP
>> Initiation Message TLVs" registry; is the section heading of "Peer Up
>> Information TLV" appropriate?
>> 
>> There is a draft undergoing to split Initiation Message and Peer Up message 
>> TLV registries, see 
>> https://tools.ietf.org/html/draft-scudder-grow-bmp-peer-up-00 , which makes 
>> a lot of sense and is being adopted by the WG. I look for suggestions on 
>> what would be the nicest way to address this.
> 
> I  think the responsible AD can handle this with the IANA interactions
> during the publication/AUTH48 process.  Thanks for letting me know about
> the other work in progress!

Ack, thank you for the suggestion.

Paolo


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

Reply via email to