On Thu, Aug 16, 2007 at 12:12:11AM +0100, Michael Rogers wrote: > vive wrote: > > Not that I'm aware of. The current plan is to see if it appears under a > > network running Opennet (path folding) with added churn (nodes joining for > > a while taking a randomized position, participating some in swapping, then > > disappearing for good). > > What's the reason to think that churn causes clustering? I can see how > #freenet-refs would create clusters of newbies, but it's not obvious to > me that churn would have that effect, on either darknet or opennet.
It's not obvious but a hypothesis we should test for. There is also a plausible explanation how it may work that has been brought up several times before to describe what is seen in the network. The same thing was also pointed out recently by the gnunet guys. The reasoning goes like this: there is an inflow of newbie nodes that start with randomizing their positions, and there is a set of nodes that is a well-connected neighborhood that has been swapping long enough to be in a fairly "stable" state (specializing in a certain part of the keyspace). Note that this is a natural part of running the swapping algorithms and having some tight-connected neighborhoods. Once in a while the randomized position from a newbie node will fit well enough with the later group, so some nodes in that cluster will want to consume (with swapping) the newbie addresses if it lets them specialize even further (brings them closer in keyspace, due to the nature of the swapping algorithm). Where a newbie has not specialized much in the network, it will tend to swap away this position with the nodes that claim to be in better need of it. The final part of the argument is that some newbie nodes have a short lifetime, so the idea is that if they disappear from the system (for good) after having swapped and participated for a while, they bring with them both data and addresses from a sparser part of the keyspace. Result: this churn adds to specialization in certain parts of the keyspace. > > Simulating a darknet-style network about if this is happening is more > > complicated, I don't know of a straightforward way to simulate churn where > > nodes leave for good in a darknet and with new ones arriving being placed > > well in the topology (with no path folding). > > Could you place the new node at a random location and then connect it to > each existing node with probabilities given by the Kleinberg > distribution? (Or a more realistic social network model... but which one?) Using the current locations is not the same as generating the Kleinberg model from the beginning. One trick may be to remember the original address when generating the network and approximate the process by adding nodes later on, another idea may be to gradually "awake" nodes and let them live for a short while. I dont know about any model that works on the current addresses. Vilhelm -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 187 bytes Desc: not available URL: <https://emu.freenetproject.org/pipermail/tech/attachments/20070816/11d65806/attachment.pgp>
