RFC bandwidth, packet size and latency

We analyzed the average page count and yearly production for RFCs. Note that there are 207 RFCs, published mostly through 1984, that are not online or do not have page numbers in an easily parsable format.

From the IMR, we also tabulated the number of Internet Drafts published, including revisions.

Analysis: The page count has fluctuated somewhat, settling on a range of 20 to 30 after 1979. A few very large RFCs can distort this metric considerably; a median would be more useful, but this is left as an exercise to the reader.

The overall RFC editor bandwidth increased dramatically after about 1985, but has fluctuated significantly between 4,000 and 8,000 pages per year in more recent years.

The data was extracted from the Internet Monthly Reports (IMR) and the IETF announcement mail archive. (For some reason, the IETF web site only has IETF announcements from August 1998 onwards. IMRs are available starting 1988.) The data was parsed and imported into EDAS and netbib, written by the author.

The publication activity was tabulated via

select
year(published),sum(pages),count(pages),round(sum(pages)/count(pages),1)
from paper where trtype='rfc' group by year(published);

The latency was tabulated via

select year(published),count(*),avg(to_days(published)-to_days(modified))/30 from
paper,paperlog where paper.paper=paperlog.paper and trtype='RFC' and
value like "%-00" group by year(published);

The raw data is available as a comma-separated file as rfc-data.txt, generated as

select number,published,modified,value from paper,paperlog where
paper.paper=paperlog.paper and trtype='RFC' and value like "%-00" order
by number,value into outfile '/tmp/rfc.dat' fields terminated by ',';

The columns are: RFC number, date RFC published, date of -00 draft, name of -00 draft.

Below is the distribution across all RFCs measured, tabulated in 3-month units.

Year I-Ds published RFCs published (total) RFCs published (with pages) Total RFC page count Average RFC page count RFCs measured Delay -00 to publication, months
1968 1 1 7 7.0
1969 20 12 50 4.2
1970 57 49 320 6.5
1971 169 153 793 5.2
1972 113 105 442 4.2
1973 158 114 625 5.5
1974 58 25 151 6.0
1975 24 11 48 4.4
1976 11 9 85 9.4
1977 19 9 90 10.0
1978 8 3 23 7.7
1979 6 3 80 26.7
1980 15 8 267 33.4
1981 27 13 482 37.1
1982 34 13 165 12.7
1983 44 38 468 12.3
1984 38 34 865 25.4
1985 39 39 645 16.5
1986 23 23 551 24.0
1987 41 41 1422 34.7
1988 45 45 1063 23.6
1990 55 55 1904 34.6
1991 148 91 91 2174 23.9 7 5.2
1992 363 94 94 2176 23.1 29 7.2
1993 567 170 170 3748 22.0 63 9.5
1994 655 176 176 4452 25.3 91 10.0
1995 988 126 126 2652 21.0 81 10.8
1996 1380 170 170 3337 19.6 129 13.2
1997 2200 187 187 4798 25.7 114 16.5
1998 2316 227 227 5924 26.1 189 21.5
1999 2523 255 255 7671 30.1 248 21.4
2000 3194 278 278 7670 27.6 276 24.0
2001 3638 190 190 4648 24.5 206 18.6
2002 3328 218 218 5754 26.4 222 23.2
2003

These statistics have a number of possible problems. They are based on matching names and draft titles from the Internet Monthly Report (IMR) with the RFC titles. Sometimes titles and I-D identifiers change as the I-D progresses from -00 to RFC, so that a precise match would require a manual analysis. (Unfortunately, there seems to be no public database for all RFC announcements; only those from August 1998 onwards seem to be available. There does not appear to be another database that records the original I-D.) Most of the early RFCs were not published as Internet Drafts, thus, we omit years before 1991.

In some cases, e.g., in 2001 and 2002, there are more entries than published RFCs. This is caused by RFCs that have multiple I-D sources. Unfortunately, there is no automated way to eliminate this error. It also tends to lower the estimate. However, just restricting this to draft-ietf-* I-Ds changes the 2002 delay to 23.6, i.e., makes hardly any difference.

Since many drafts change title or identifiers (from draft-person to draft-ietf), the delays are likely to be conservative.

Since the IMRs do not record the precise publication date, we assume that each I-D is published in the middle of the month. Months are measured at 30 days.

We only include RFCs where we have a date for -00 draft. For reasons that are unclear, some drafts do not appear in the IMRs.

Some remarks:


Internet Technical Notes and Resources

Last updated by Henning Schulzrinne