A Totally Unscientific Study of Self-Similarity

I run a moderately busy WWW, FTP, and mail server. It's also used for interactive logins and occasionally acts as a fileserver. I started a script running that would take a reading every second of the number of packets by my machine transmitted in the previous second. This was trivially done under Linux simply by running running "ifconfig eth0" and looking at the "TX packets:" line.

Next, I wrote a couple of simple scripts to average and plot the data over 1, 10, 100, 1000, and 10,000 second intervals.

I was struck by how... well, similar they all looked. It looks like there are bursts at every time scale.

I don't have any claims about what these data mean, except that it's not hard to find anecdotal evidence of "visually self-similar" traffic. And that graphs are pretty.

Back to my writings page
Back to my home page

Web page last modified on 5 May 1999