In this section, we first evaluate the performance of P2Cast
through simulation experiments. We then
compare three overlay
construction algorithms: BF, BF-delay, and
BF-delay-approx. In our simulation results, the half-width of the 95%
confidence interval of the
data shown in this paper is always less
than 5% of the point estimate.
The experiments show that (1) P2Cast is more scalable than
either a client-server unicast approach, or an IP
multicast-based patching approach; (2) the
larger threshold helps to serve more clients in
P2Cast; and (3) under the same conditions, BF-delay and
BF-delay-approx algorithm can serve more clients than BF and
reduce the overall network workload over BF. However, they
present a higher workload to the server than BF.
We will address the failure
recovery problem in the Section 5. For now we
assume that no client departs early and that there are no network
failures. We start with the introduction of the simulation setting.