Good, large-sample data on Internet traffic is hard to find. However, an excellent study of Internet traffic in Japan [Cho et. al. (2006)] describes 42% of Japanese (public) Internet traffic for six one-month observations spanning Sept. 2004 to May 2006.
This study reports many important facts. The ratio of bytes sent to residential customers to bytes received from residential customers was 1.3 — remarkably symmetric. Peer-to-peer applications using dynamic port assignment account for most of the traffic. The distribution of bandwidth use by users is a high-powered power law. Most importantly, specific users’ positions in that distribution vary over time and are not well identified with demographic and other customer characteristics (other than fiber connection, i.e. bandwidth availability).
The residential traffic growth figures for Japan vary considerably across month intervals. From Sept. 2004 to Nov. 2004, total traffic (aggregated inbound and outbound) grew 180% on an annualized basis. In subsequent 6-month intervals, annualized traffic growth rates were 58%, 19%, and 37%. Data on growth in fiber-to-the-home (FTTH) connections and total connections also shows volatility, but this does not seem to explain the reported traffic growth volatility. Any ideas about what explains it?
According to my calculations, (public) Internet traffic in Japan from Sept. 2004 to May 2006 grew about 60% per year. This figure aggregates inbound and outbound traffic and includes residential broadband customers and non-residential broadband customers (leased lines, data centers, dialup). I estimated non-residential traffic for the 7-ISP sample using the reported residential/non-residential traffic figures from the 4-ISP sample in Nov. 2005, and the non-residential broadband growth rate for the 4-ISP sample from Sept. 2004 to May 2006.
This estimated annualized growth of Internet traffic in Japan is much higher than estimated annualized growth of (non-voice channel) bandwidth in use in the U.S. across the 1990s. From 1989 to 1999, total DDS, DS1, and DS3 channel termination bandwidth in the U.S. grew an estimated 27% per year (see Table P6 in U.S. Bandwidth Price Trends in the 1990s). The difference between a 60% growth rate and a 27% growth rate year becomes huge after only a small number of years.
The Japanese study does not encompass bandwidth deployed in private networks. The ratio of Internet bandwidth to total inter-office bandwidth may have been about 15% in the U.S. in 1998 (see Growth in the “New Economy”, p. 6). The Japanese data show much a faster growth rate of non-residential bandwidth than residential bandwith, but the former is only about two-thirds the size of the latter. This is consistent with a large share of non-residential bandwidth not being incorporated into the public Internet.
An important point: the re-organization of network transmission protocols on existing networks tends to occur very slowly. Astonishing fact: in the U.S., about 90% of mobile-phone communications towers use traditional, copper-based TDM backhaul. Legacy networks hang around for a long time.
Reference:
Cho, Kenjiro, Kensuke Fukuda, Hiroshi Esaki, and Akira Kato, “The Impact and Implications of the Growth of Residential User-to-User Traffic,” sigcomm 2006.