[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: snmpconf Comments on the BCP

At 03:18 PM 9/28/2001 -0700, David T. Perkins wrote:

First I read draft-ietf-snmpconf-bcp-06.txt.
Then I read Dave's comments on -05.

I have to agree with Dave on many points, especially the general comments.
I think it would be difficult (at best) for a non-expert to easily 
understand how
to better design or implement any component of an SNMP-based management system,
after reading this BCP.

I think sections 2 and 3 should be rewritten to improve clarity and focus.
There are many pearls of wisdom buried in way too much text. There
are obscure examples, 'dangling examples' that are unexplained, and
unanswered questions raised throughout these sections.  However,
these 2 sections have a lot of potential value to the reader, and
also contain the kind of info that actually belongs in a BCP.

Section 4 could also use some major edits for clarity and focus.
There seems to be a lot of text about MIB design in some sections,
and there is no mention of get-next processing, multi-phase
set PDU processing, multi-PDU set handling or counter caching issues
(for starters).  Section 4.5 is not really true.  The order of the varbinds
in the PDU is not important to agents.  It's not mentioned in the I-D,
but the presence (or absence) of related object instances in the same PDU,
and the presence of partial and/or multiple instances of related
objects in the same PDU, is quite important to the agent.

Section 5 seems reasonable, but I'm not an SNMP application developer.
There are many aspects of NMS design not mentioned at all.  The
BCPs for set processing (5.3) focus on transfer efficiency and do not even
mention the impact the varbinds present (or not present) in the Set PDU
have on agent complexity.  This section could say that applications should
avoid use createAndWait if at all possible and always include all related
object instances in the same PDU. NMS developers should be aware that
agents may limit the number of related instances created in the same PDU
(as well as overall).

Section 6 starts with a procedural model that is hard to understand in the
current table format. It should be rewritten as simple paragraphs for clarity.
The DOCSIS access control example (6.4) seems like overkill to make
the point that SNMPv3 security mechanisms should be used.
There should be a lot of BCPs related to VACM and USM that could be
added by experts in those areas.  I think security deployment is the
most important and relevant BCP topic this document addresses,
but there's nothing about how to deploy SNMPv3 security.

Sections 7, 8, and 9 are tightly coupled to the PM MIB and
do not really represent BCPs.  The I think these sections should be
moved from the BCP to the PM (or other) document. Section 1
talks about how this document represents a decade of experience.
That is hardly the case with the concepts and PM-MIB specific
examples described in these sections.   There could be some
applicable BCP nuggets in here,  but they are buried in a lot
of descriptive text about the PM MIB.  It's not clear to me what BCP
the entire Heating and Air Conditioning MIB example is conveying.

I think this document needs significant changes.
The -06 version is not ready for RFC publication as a BCP.


>The following URL points to comments on the -05 version
>of the BCP. I believe that most are still appropriate,
>except some of the grammar, spellings, etc are cleaned
>up in the -06 version. However, I haven't read it,
>and don't plan to.
>   http://www.snmpinfo.com/ftp/temp/comments-05.pdf
>And it is OK (and encouraged for the document to be
>copied to the SNMPCONF repository).
>The bottom line on the document for me is that it needs
>extensive work before it is approapriate to be published.
>It is troubling that the WG chairs and authors would
>even consider forwarding the document for publication
>in its current state.
>I hope that the document is rewritten because there
>is a need for it.
>/david t. perkins