Simulation setting

We use a 3-level network topology in our simulation experiment. Fig. 3 illustrates the top two levels generated by GT-ITM [12] with 100 nodes. We assume that each node is an abstraction of a local network that can host an unlimited number of clients, and there is sufficient bandwidth within a local network to support media streaming. The network consists of one transit network (consisting of 4 nodes) and 12 stub domains. The shortest path algorithm is used to determine the routing. We assume that the video playback rate is constant bit rate (CBR). We assign a bandwidth to each link in terms of the number of playback rate a link can support. The capacities of links between transit nodes and between transit nodes and stub domain nodes are chosen to be larger than those between stub domain nodes, since links in the core network are typically better provisioned and have more bandwidth than the edge links. Using advanced coding techniques, videos with the playback rate from 300bps to 500bps offer reasonably good viewing quality. A link with the capacity of 100Mbps can support 200 to 333 such streams. Since we simulate P2Cast providing service for one video, we choose the capacity of each core link to be 20, i.e., core links can support up to 20 streams simultaneously; and that of each edge link to be 5 for the simulation results reported in this paper. Furthermore, we change the location of server from the transit network to the stub domain to study the sensitivity of the performance to the server bandwidth change. We also vary the link capacity, and similar results are observed in [11]. We simulate the on-demand service of one video to clients whose arrival process is Poisson. Each client is equally likely to be placed at any node. It is possible for more than one client to reside at the same node.

Figure 3: Top 2-level network topology used in simulation.