what-when-how
In Depth Tutorials and Information
random node failures or departures (churn) are quite common, this resilience is
necessary to maintain eicient sharing of the ile by all nodes in the system. his is
especially important for the Unchoked network generated by the incentive mecha-
nism, as node failures should not result in a decreased usage of the system's total
uploading or downloading bandwidth. he tolerance of scale-free graphs to node
failures has also been shown to extend to misbehaving nodes in the system. his is
also desirable in BitTorrent to avoid performance problems that could occur due to
free-riders in the system.
As described in Section 1.3.2, we will determine if a network is scale-free by
doing a linear it to the node degree plot on a log-log scale. A sample node degree
distribution and it is shown in Figure 6.11.
Figure 6.12 shows the
R
2
goodness of it values for the in-degree of the four
networks throughout the experiment. he only network of the four that exhibited
this power law behavior was the Unchoked network, which had an
R
2
goodness
of it value of approximately 0.9 over most of the experiment (except during the
startup stage). his is high enough to indicate a good it, while the other networks
had goodness of it values less than 0.7.
Figure 6.13 shows the power law exponent found from the fitting of the in-
degree of the nodes in the Unchoked network. he power law exponent can be seen
10
4
Raw data
Fit (γ = 2.07 ; R
2
= 0.87)
10
3
10
2
10
1
10
0
10
−1
10
0
10
1
10
2
Node Degree
Figure6.11
Thenodein-degreedistributionfortheUnchokednetworkathour
19oftheexperiments,andtheresultingittoit.
