evident from the values of Φi; such resources are exploited in a random way by the other parties using the fixed probability
criterion for neighbor selection. Recalling that in Table 3 the reported values are averages computed when the P2P system
reaches its equilibrium, one notes that the P2P application provides an unfair resource usage. In fact ISP-1, that has the
largest upload, provides on average Ψ1 = 255.10 × r chunks regardless of the fact it requests only Θ1 = 50.31 × r chunks
to serve the 10% of the peers in the swarm. On the other hand ISP-3, that is the largest in terms of size but has a limited
upload, requests Θ3 = 397.14 × r chunks whereas it contributes only with Ψ3 = 71.80 × r chunks. The effect of EGT
can be appreciated on the right hand side of Table 3. One observes that the EGT algorithm is able to equalize the requested
and offered resources (see the values of Θi versus Φi) pushing the system towards the optimal point identified by Eq. (4),
that clearly corresponds to a criterion of effort fairness. Also in this case the offered resources are exploited according to the
rules of the P2P application determining the values of the upload that are actually provided. It turns out that by means of
EGT, ISP-1 is able to reduce its upload towards the other ISPs down to Ψ1 = 48.47 × r chunks. To counteract this ISP-3 is
asked to increase the resources it provides from about 72 × r to 86 × r chunks. Finally, it is worth pointing out that the EGT
optimization does not incur any penalty in terms of the offered service leading to values of QoSi
larger than 0.98 in all the
simulated cases.
Table 4 shows the results obtained by changing the behavior of the P2P application, that in this case implements the
variable probability neighbors selection criterion (p = mi/
∑3
j=1 mj). The obtained results are quite similar to the one we
have just commented except for the absolute values of the requested and provided resources, that clearly depend on the
efficiency of the P2P application. As an example, in absence of EGT, the variable probability criterion turns out to be less
unfair for the ISP-1 (Ψ1 = 81.86 × r chunks) with respect to the results in Table 3. Nonetheless, the EGT is again able to
equalize the resource usage leading to a lower Ψ1 = 70.74 × r chunks without impacting on the quality of service.
From previous analysis of the experimental results we can conclude that the EGT improves the resource usage fairness but
does not affect the quality of the service, provided that enough resources can be allocated to the service by the ISPs so as to
guarantee service continuity. Moreover, we have shown that the EGT does not depend on the neighbor selection criterion, thus
turning out to be useful independently of the approach adopted by the P2P application to build and maintain the streaming
overlay.