ietf
[Top] [All Lists]

Re: Comments on draft-carpenter-newtrk-questions-00.txt

2006-07-13 11:59:27
Has this been exercised in the past, say, 5 years?

At least for widely-implemented protocols, such as SIP, there is no reasonable way to say "there is only one implementation that does X", as there is no plausible way to catalog all such implementations. Most of the implementors don't show up at IETF meetings and implementations are written by dozens of small start-ups, open source systems and other non-traditional sources.

This provision actually discourages any DS effort: the danger is that you can't find an implementation that does use a certain feature, you deprecate it and then find that there was some SDO or implementor somewhere that actually did find this extremely useful. That just makes everyone look silly.

On Jul 13, 2006, at 7:29 AM, Joel M. Halpern wrote:

there is one useful aspect of our DS contortions. When we do the work, we actually figure out which features turn out not to be used, and get them out of the spec. OSPF had ToS routing in its PS specification. It turned out that there was only one implementation.
So when the protocol was ready to advance, that feature was removed.
Knowing that the feature was not being used proved very helpful to us later.

Yours,
Joel

At 02:45 AM 7/13/2006, Eliot Lear wrote:
Fred Baker wrote:
> I would like to believe that a well documented interoperability test
> constitutes DS qualification; the current DS qualification sets the
> bar somewhat higher than that, and I note that few documents actually
> achieve that, even though we can daily see implementations
> interoperating in the field at PS.

Some data to Fred's point:

By RFC, not by STD (obviously):

Status  1999    2000    2001    2002    2003    2004    2005
-------------------------------------------------------------
PS      102     119     71      105     103     131     169
DRAFT   6       6       2       4       7       7       3
STD     3(*)    2       0       8*      3       0       1


(*) 3 in 1999 were SMIv2 6 in 2002 were SNMP.

These are rough based on 10 minutes of scripting I did back in March. I believe there are two reasons for the huge gap between PS and DRAFT:

 - it's difficult to get there (interop requirements, picking out
   uncommonly used features, etc)
 - nobody wants or needs to do the work (what GM in her right
   mind would want her experts working on something that neither
   generates new features nor fixes product bugs)

If Iljitsch's proposal is that the IESG "makes a call" based perhaps on somebody's request with some modest effort to demonstrate that a spec is ready for the next step, I think that actually would be a fine two-step approach.

Eliot

_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/ietf


_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/ietf


_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/ietf