Summary: The Impact and Implications of the Growth in Residential User-to-User Traffic

This paper looks at some high-level Internet traffic measurements for six of
Japan's major ISPs. What makes this study unique is that it appears they
were able to look at a substantial portion of the country's traffic.

The main thrust of their work seems to be looking at what the effects of
residential fibre optic connections will be on their national backbone. The
paper seems to imply that ISPs may need to either enlarge their backbone
connections or rethink their residential pricing models if bandwidth usage
continues to increase.

Like the UW paper, these logs also show that a few users (4%) use a
disproportionately large (73% inbound) portion of the bandwidth. However,
the authors point out that no real dividing line can be drawn between the
heavy-hitters and the regular users. The authors claim that the wide
distribution of heavy-hitters suggests that they don't in fact form a
separate class -- they don't use any exotic applications, they just use them
more often. This means that total bandwidth use will likely continue to rise
rapidly as today's "normal" users simply do more video-downloading, etc.

The paper also notes some interesting geographic and temporal properties that
would be difficult to capture without conducting a trace on a national scale.
For example, the authors claim that the bandwidth consumed per prefecture is
proportional to the prefecture population, without regard to how rural or
developed it is. The paper also notes that there are proportionately just as
many "heavy-hitters" in rural areas as there are in urban prefectures.

I thought the data in this paper was very interesting. It would be very
interesting to see a study with as much detail as the UW paper, but on a
national scale. A national-level trace of US traffic would be very useful,
but it would probably be quite difficult to get ISP buy-in.
Received on Tue Nov 21 2006 - 03:14:00 EST