Figure 8 Average and maximum node connectivity (in the biggest connected component)
Related Figures (13)
Figure 3: Graph size and diameter (data collected over a three week period, 14 measurements) We ran the crawler to generate graph data during a one-month period. As you may see from the following figure there is a steep increase in network size after November 20" We believe this is not due to a change in Gnutella’s network but we link it to an improvec crawler version (we built a distributed crawler that is much faster). We see that the diameter, which is the longest short distance between two nodes, always fells to smaller values, such as 7 or 8. The reason for this might be that Gnutella servants typically issue ping messages with TTL=7. Figure 4: Aggregated size of data shared (data collected over a three week period, 14 measurements) Figure 5: Number of files shared by each node (data collected over a three week period, 14 measurements) Figure 10: Usage patterns: data shared against node connectivity There are a couple of things that we observed during this experiment: Considering a 40 bytes TCP/IP headers overhead per message and ignoring the overhead brought by lower network layers the next plot presents the overall volume of date transferred (in Kbps) averaged over one minute intervals (general average for out 6:15 hours measurement is 11.5Kbps). We observed a high number of hosts that were reported as members of the network although we could not contact them. The next figure details the reasons for which we couldn’t contact those servers. For most of them it is probable that the crawler tried to investigate them after the node left the network. However there was a significant percent of hosts that did not follow Gnutella’s peer to peer connection establishment. 4.10 Domain analysis
