Carter,


And the place where you and I part company is the emphasis placed on “process” 
(e.g., “the only relevant IETF process document for the NVO3 WG”) - I’m 
interested in seeing this WG deliver results ... both the WG charter and 
requirements documents are tools that can contribute to achieving that goal, 
but the focus should be on the goal, not the tools used to achieve it, IMHO.



Thanks,
--David (chair or past chair of 5 IETF WGs)



From: Carter Bullard [mailto:[email protected]]
Sent: Saturday, February 07, 2015 1:15 PM
To: Black, David
Cc: [email protected]
Subject: Re: [nvo3] Proposed timeline for publishing Requirements



Hey David,

Its pretty simple … Reading only from the NVO3 charter, which I believe is the 
only relevant IETF process document for the NVO3 WG, and skipping to the 
paragraphs that describe how the protocols will be developed....



   The NVO3 WG may produce requirements for a network virtualization control 
plane, and will select, extend, and/or develop one protocol for each of the 
functional interfaces identified to support the architecture.



   I interpret that to mean that a requirements document will identify 
functional interfaces, from which one protocol will be selected, extended 
and/or developed.  I think that document should be an RFC.

   The NVO3 WG may produce requirements for network virtualization data planes 
based on encapsulation of virtual network traffic over an IP- based underlay 
data plane. Such requirements should consider OAM and  security. Based on these 
requirements the WG will select, extend, and/or  develop one or more data plane 
encapsulation format(s).

   I interpret that to mean that a requirements document will be used to 
select, extend, and/or develop one or more data plane encapsulation format(s).  
I think that document should be an RFC.



   My only point, is that if you do not take whatever requirements documents 
you have to RFC, then you will not have completed the dependent task defined in 
the WG charter for developing NVO3 protocols.  Yes you have an architecture 
RFC,  great, first step in the charter.  Next step is “may produce 
requirements”.  Great, but if you move on to “select, extend and/or develop” 
you need to have formalized the requirements to the point where you can 
identify the functional interface.  That I believe would be an in the form of 
an RFC.



   I hope you can understand my position.  It is simply a superficial 
interpretation of the WG charter, and why I suggest that you should take any 
requirements documents that the WG would like to use, to RFC.



   Carter







   On Feb 7, 2015, at 12:52 PM, Black, David 
<[email protected]<mailto:[email protected]>> wrote:



   Carter,




   Maybe you should go back further.  How about say RFC 985, 1009, 1716, and 
1812.
   Four RFCs devoted solely to trying to get the requirements for routers down.
   At a time when some of the best work was done.


   Perhaps you should think more about the types and contents of requirements
   documents.  Those four router requirements RFCs contain *implementation*
   requirements for routers, and analogous implementation requirements are
   to be found in almost every IETF protocol specification.

   In contrast, the NVO3 requirements documents (and I'm an author of some
   of them) are protocol *design* requirements documents; they do not contain
   implementation requirements (those go into the protocol specs).  There is a
   difference, really there is ...

   Just for grins, I dug up a design concerns draft of mine (it includes
   design requirements) from over a decade ago:

             http://tools.ietf.org/html/draft-black-rdma-concerns-00

   That draft accomplished its intended purpose of influencing the WG's design
   activity; in that case, publication as an RFC was not needed (and in fact
   the draft even said so up front).

   Thanks,
   --David




   -----Original Message-----
   From: nvo3 [mailto:[email protected]] On Behalf Of Carter Bullard
   Sent: Friday, February 06, 2015 4:09 PM
   To: Melinda Shore
   Cc: [email protected]<mailto:[email protected]>
   Subject: Re: [nvo3] Proposed timeline for publishing Requirements

   Melinda,
   Maybe you should go back further.  How about say RFC 985, 1009, 1716, and 
1812.
   Four RFCs devoted solely to trying to get the requirements for routers down.
   At a time when some of the best work was done.

   The original charter for NVO3 states that optional requirements documents 
could
   be generated, from which solutions would be implemented.  The WG decided to
   go down the optional route, in order to go after solutions.  The original 
charter
   stated that these solutions would come from requirements documents.  I 
interpret
   that to mean that the requirements would be IETF documents, which I 
interpret to
   be RFCs.

   The only issue is, if you can proceed successfully, then proceed.  Experience
   suggests that formal goals help a WG to be successful.  Httpbis had simple
   requirements, catalog security properties and provide improved bindings of
   semantics to underlying transport, no need for a requirements document there.
   Not like it needed to change an application protocol much to be successful,
   and
   the charter left a lot unspecified to allow for whatever else anybody wanted
   to
   throw in.  No mention in the charter that the WG would work through
   requirements
   documents to reach solutions.

   If you think NVO3 is an equivalent to Httpbis, then I'm at a disadvantage, as
   I'm not
   that good at thinking at that level.

   Its up to the WG to decide if it can proceed successfully without formal
   requirements.  My opinion is that there are a lot of ' time to market '
   pressures
   that will compel implementations to be less than adequate, and without
   good control over the design goals, these implementations will get leverage
   over the market that the WG is trying to influence.  Which would be a shame.

   Carter






   On Feb 6, 2015, at 3:35 PM, Melinda Shore 
<[email protected]<mailto:[email protected]>> wrote:

   On 2/6/15 11:15 AM, Carter Bullard wrote:



   Hey Melinda,
   If you're point is that you have to go back to 1999 to find an IETF

   protocol that



      was successful, then seems that you're in agreement with my point.


   I'm not.  HTTPbis doesn't have a requirements document, either,
   nor do a large number of other successful protocols.  Arguably
   the IETF produced more useful, relevant documents back when
   requirements documents and whatnot were not the norm.

   I think that in general requirements documents are helpful to
   the process of specification and have some historical value, but
   at this point it's very clearly the case that the IETF process
   has slowed down to the point that it's gotten nearly impossible
   to publish documents in a timely way.  It's happened at the same
   time that documents that used not to be part of the process -
   problem statements, requirements documents, gap analyses, and
   so on - have become part of the culture.  There's nothing in
   any IETF process document that requires them.  Allow me to repeat
   that: there's nothing in any IETF process document that requires
   them.  So, to the extent that they're useful, yay, to the extent
   that they bog the IETF down in non-specification work, boo.

   Melinda

   _______________________________________________
   nvo3 mailing list
   [email protected]<mailto:[email protected]>
   https://www.ietf.org/mailman/listinfo/nvo3


   _______________________________________________
   nvo3 mailing list
   [email protected]<mailto:[email protected]>
   https://www.ietf.org/mailman/listinfo/nvo3



_______________________________________________
nvo3 mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/nvo3

Reply via email to