Stringency (RE: 12 problems)

[Harald Tveit Alvestrand]

--On torsdag, januar 16, 2003 10:52:49 -0800 James Kempf 
<kempf at docomolabs-usa.com> wrote:

> Whoops! Cancel that analysis. If the number is "months per RFC", then it
> actually showed an *improvement* between 1995 and 2000.

I'm not sure I can manage to interpret that number...

> 1993    0.177
> 1994    0.105
> 1995    0.172
> 1997    0.139
> 1998    0.133
> 1999    0.126
> 2000    0.153 (note: I changed this - I think you had it wrong..)

if the IETF had been a single person, devoting its attention to one RFC at 
a time, the numbers would indicate that the IETF needed 0.126 months of 
attention to produce an RFC in 1999 (taking 19 months to produce 157 RFCs) 
, but 0.172 months of attention to produce an RFC in 1995 (taking 10.5 
months to produce 61 RFCs)?

This of course ignores the fact that most of the work in producing the year 
2000 RFCs was done in 1999, and certainly impacted work on the 1999 RFCs, 
which were mainly produced in 1998, and so on back the chain....

besides, what does it mean for an entity to work 19 months in one year?

If instead we *multiply* the number of RFCs by the time needed to produce 
them, we should get something indicating the amount of attention directed 
at the IETF doucment-producing effort:

Year	RFCs	Delay	Effort	Largest IETF Effort/attendee	Watchers
		months	RFC*mn		      RFCmonth/person
1993	31	5.5	170.5	638 (Columbus)	    .26		44.9
1994	80	8.4	672.0	1079 (San Jose)	    .62		19.2
1995	61	10.5	640.5	1007 (Dallas)	    .63		18.8
1996	96	12.4	1190.4	1993 (San Jose)	    .59		20.1
1997	103	14.4	1483.2	1897 (Washington)   .78		15.3
1998	147	19.6	2881.2	2124 (Orlando)	   1.35		8.8
1999	157	19.8	3108.6	2379 (Washington)  1.30		9.1
2000	158	24.2	3823.4	2810 (San Diego)   1.36		8.8

there may be SOME evidence for the "adding more people to a late project 
makes it later" theory - but the main point may be that unless we know what 
theory we want to test, it's a bit dangerous to multiply things together...

....for another interpretation of the numbers, you can invert the 
effort/attendee, multiply by 12, and call it "number of people looking at 
each document being written", and argue from there that our problem is that 
we have too *few* people pushing at each individual document.... in 1993, 
there were on the average 14 RFC-projects going on, each with 45 people 
watching it; in 2000, there were 318 projects going on, each with less than 
9 people watching it.

Of course, this probably has *very* little to do with root causes or 
remedies - if we were to keep the same attendance, but could finish 
projects faster, we would have more eyeballs free to look at the documents 
we *are* working on.

But still, it'll be fun to look at what these random multiplications do to 
the 2001 and 2002 numbers :-)


                   Harald