RFC bandwith, packet size and latency

[Harald Tveit Alvestrand]

Again thanks to Henning for making these data available!

for some reason the .txt file contained 1697 lines, which didn't seem to 
get all the RFCs that Henning counted, so the numbers don't quite match 
up.... but I did some other calculation - looking at average delay by when 
projects *started*.

The thinking behind this was that we have gotten some old things out the 
door recently (DHCPv6, AAA and IPv6-addr-arch comes to mind) - if 
summarizing by year of publication, finishing an old project will make the 
statistics for that year look worse, while most of the errors that led to 
the project being slow probably (?????) occured near the beginning of the 
project.
Under the theory that people tend to optimize for what is measured, this 
would encourage the termination of old projects rather than finishing them 
- and the merits of doing that might be variable.

The two sets of numbers:

Delay by year of publication
1991 7   5.2
1992 29   7.2
1993 62   9.3
1994 82   9.4
1995 71   9.6
1996 105  12.1
1997 99  15.1
1998 149  17.8
1999 215  19.9
2000 252  20.6
2001 172  18.0
2002 204  22.5
2003 16  17.8

Delay by year of version 0
1991 41  14.1
1992 54  14.7
1993 69  12.9
1994 113  16.6
1995 121  18.6
1996 145  18.7
1997 230  20.2
1998 179  19.1
1999 217  18.2
2000 157  15.7
2001 108  12.6
2002 29   7.6

This seems to indicate that 1997 was the year where we started by far the 
most projects that have so far concluded, and that it was (unsurprisingly) 
also the year with the longest-running projects.
But the change in time-to-publish seems to be much more gradual than what 
the time-to-publish by year of publication indicates.

Of course, the numbers for 2001 and 2002 are more or less meaningless - a 
lot of projects started in these years will not yet have concluded.

Comments?

                          Harald