Dr. Kenjiro
Cho is a renowned researcher of the
Internet in Japan. His famous paper on The Impact and Implications of the
Growth in Residential User-to-User Traffic was the first to point to the
real observed usage of high speed broadband by a large population.
In my words: at one point in time 5% of the population is a heavy user,
but its never the same 5 %. People use the capacity casual when needed and
expect an immediate response. Just like we use our electricity network, water
network, sewer network. They are not built for averages.
Think of sewers pipes from your home, what a mess would it be if they were designed for the average flow and you try to flush your toilet.
At the IETF meeting in
Japan last November on "Internet Bandwidth Growth: Dealing
with Reality" Dr. Kenjiro Cho talked about his latest findings (
He used some statistical terminology in his talk that needs some explanation, lest people draw the wrong conclusion. Dr. Cho was kind enough to clarify the data for me, which I would like to share. [Update] The interpretation is mine.
The first data point is the gradual shift in average residential data consumption in Japan.
In the period 2005 -2009 the average upload
grew by 29 %, the average download however grew by 117 %. The symmetry ratio's (up:down) changed from 1:1 in 2005 and 1:1,7
in 2009. The main reason is the increase in streamed video.
But
averages can be misleading (as in the case of your sewer pipe when you flush the
toilet a couple of times per day). An indicator of the variation is the
difference between the mean (average) value measured and the mode value of the measured
distribution. (The mode value is the value that has the highest count in the table of measurements. For
instance the row (400,1,1,1,300,1,2,1,350) has a mean value of 117 and a mode
value of 1).
The bigger
the difference between mode and mean, the more variation there is in the measured values.
The ratio
of observed mean versus mode values are
123:1 in 2005 and 93:1 in 2009 for the upload, indicating a very high
variability in usage patterns for the upload. Sometimes some people are sending
large amounts of data.
For the download
the ratio is 14:1 in 2005 and 8,5: 1 in 2009, less variation than observed in the upload and consistent with the observation that a lot
of streamed video is viewed.
Dr. Cho's
observations show that bandwidth usage in ultra high speed broadband networks
is consistent with a casual usage
pattern with very high variations, like we use other utilities.
And that
averages can be very misleading.
Leave a comment