Most research churn pattern is done with Gnutella network. 1. http://www.imconf.net/imc-2006/papers/p19-stutzbach2.pdf 2. http://www.cs.uoregon.edu/~reza/PUB/mmcn06.pdf 3. http://prisms.cs.umass.edu/brian/pubs/chu.labonte.p2pjournal.pdf
According to those, you can reasonably approximate the session time of peers with a 2-pieced power-law probability distribution. And you can work out the leaving rate from that. However, networks like Emule or Bittorrent will have a slightly different pattern. Especially in BT, almost all users would stay until complete their download and then leave. There's not many studies about churn in such systems. Anh. hotbaby wrote: > Hi,all > > Extract samples from all online nodes of one chosen popular P2P > application(BT, Emule or Maze) and sort them based on average online time > ranking. Then divide them into 2 nodes group,4 nodes group,8 nodes group,16 > nodes group,32 nodes group,and so on(It means that the first group has 2 > nodes, the second group has 4 nodes, the third group has 8 nodes, and so on). > I have one question if average probability of node's leaving from the system > in each group follow some rules or who can tell me if there are related > researches existing can draw the conclusions which i want. > > Thanks > winkey > > _______________________________________________ > p2p-hackers mailing list > [email protected] > http://lists.zooko.com/mailman/listinfo/p2p-hackers _______________________________________________ p2p-hackers mailing list [email protected] http://lists.zooko.com/mailman/listinfo/p2p-hackers
