what-when-how
In Depth Tutorials and Information
450
400
350
300
250
200
150
100
50
0
0
50
100
150
200
250
300
350
400
450
Peer Index
Figure6.20
Theconnectivitymatrixathour8.
departures and arrivals together.* his experiment was the same as in the previous
section on the piece population in the presence of churn. As before, the increased
churn occurs when the number of seeds decreases rapidly; for example, from 20 to 25
hours, 40 to 50 hours, and 60 to 70 hours. Since the number of peers is limited, the
periods between these increased churn periods exhibit a state of decreased churn as
compared with the previous experiment. We find that this varying churn had almost
no effect on the power law exponent or the characteristic path length of the resulting
graphs, which are identical to the previous results shown in Figures 6.13 and 6.15.
Figure 6.24 shows the clustering coefficient for the four networks in the experiment
with varying churn. During the initial stage, it is very similar to Figure 6.16, decreasing
rapidly as the peers enter the system. However, after the initial stage (i.e., after 4 hours), the
effect of the varying churn can be clearly seen on the Connection, Interest, and Download
networks, causing their clustering coefficients to oscillate. Interestingly, the clustering
coefficient increases during the periods of light churn and decreases during the periods of
heavy churn. Although the varying churn continues throughout the experiment, the oscil-
lations in the clustering coefficients of these graphs are greatly reduced after 50 hours.
*
he previous experiment also had churn, but the amount of churn was steady throughout the
experiment.
