18 months

[Brian E Carpenter]

below...

Harald Tveit Alvestrand wrote:
> 
> --On onsdag, januar 08, 2003 18:33:30 -0800 Dave Crocker <dhc at dcrocker.net>
> wrote:
> 
> > Harald> So I would suggest that when you claim a pattern, you also
> > mention at least  Harald> 3 specific examples that you think fit the
> > pattern.
> >
> > Yes, that would be nice, just as it would be nice if those who think
> > that groups taking years to produce anything useful would provide data
> > showing how wonderfully successful such long projects were.
> >
> > In lieu of that, a rough consensus of the community should suffice.
> >
> > If the community thinks we've been doing great with working groups
> > taking 3, 5 and 8 years, then that's dandy.  (Again, I admit to
> > curiosity for examples, since I cannot think of any, within the IETF.)
> 
> Hmmm.... coming up with examples of long-lived working groups is certainly
> no problem at all.....
> 
> Diffserv was chartered before April 1998, according to the Wayback
> Machine's first copy of the charter (dated Nov 98; it's got April 1998
> milestones in the charter).

First WG meeting was March 98. The BOF (fddifs) was April 97 however.
The reason the WG is still technically alive (but no longer meeting)
is that we are waiting for the PIB to come out as an RFC. It's 
very close. I'm eagerly awaiting the day I can ask the AD to shut us
down.

> 
> RFCs published:
> 
> 2474 - Dec 98
> 2475 - Dec 98
> 2597 - June 99
> 2598 - June 99
> 2836 - May 2000
> 2983 - October 2000
> 3086 - April 2001
> 3140 - June 2001
> 3246 - March 2002
> 3247 - March 2002
> 3248 - March 2002
> 3260 - April 2002
> 3289 - May 2002
> 3290 - May 2002
> 
> The working group has taken at least 4 1/2 years, but has produced its
> first RFC before 18 months (if I have the start date right), and has
> produced documents at less than 12-month intervals.
> 
> I don't think many will say that we did "great" on diffserv, but I do think
> that the specifications produced by this group have seen takeup in the
> industry.

They are in a lot of products. I think they are used a lot more in
enterprise networks than in ISP networks. But for something that Bob Braden
once described as something like "the biggest architectural change in IP 
for many years", I don't think a 5 year span is out of order. Objectively, 
we've been fine tuning since June 1999 - the basic work took about 15 months 
from the first WG meeting.

> 
> Another example, which I personally consider more fit for the "failure" bin
> (because of lack of industry pickup): SVRLOC.

Actually, I see increasing signs of SLP being prerequisite for other 
protocols, and it is finally getting supported. It's just had a long 
infancy. This case clearly tests the 18 month "rule" and finds it false.
But I have to agree, this one was sloooow.

> 
> Wayback machine charter at May 1997 shows unfilled milestones in June 1996,
> so we can assume it was chartered before then.
> 
> RFCs published:
> 
> 2165 - June 1997
> 2608 - June 1999
> 2609 - June 1999
> 2614 - June 1999
> 2926 - Sept 2000
> 
> The group has been active for at least 6 1/2 years, and has taken 2 years
> from its first RFC to its second.
> 
> Third example, which is definitely a cornerstone of our technology: IDR,
> the owners of BGP.
> 
> Wayback machine charter from May 1997 shows milestones in june 1996, but
> it's listed as owning RFC 1105, which is published in june 1989 (and refers
> to the "IWG Task Force); I'll leave the interpretation of that history to
> the people who were around at the time.
> Its charter shows some 35 RFCs; I don't think I want to list dates for them
> all - but there are no obvious long-term "holes" in the number spread.
> RFC 1654, BGP-4, was published in July 1994.
> 
> The point?
> 
> Stuff varies. Unless we dare to look at specific examples, we can't verify
> that patterns exist.

I'd go further - we have to conduct a statistical analysis to find out
at least what are the averages and the quartiles, and whether groups
in the upper and lower quartiles are more or less successful in terms
of objective measures such as product implementations.

But it is fairly clear from the facts we have that one size doesn't
fit all.

   Brian

> 
>                          Harald
> 
> [for Wayback Machine: see www.archive.org]

-- 
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Brian E Carpenter 
Distinguished Engineer, Internet Standards & Technology, IBM 
On assignment at the IBM Zurich Laboratory, Switzerland