On Wednesday 20 February 2008 14:17, Michael Rogers wrote: > On Feb 7 2008, Matthew Toseland wrote: > >> Perhaps a node should only attempt to reduce its > >> own stress (in Robert's terms), rather than the product of its stress > >> and the other node's stress? > > > >That's an interesting idea. Wouldn't it result in much slower network > >evolution? Simulate it, if it looks promising we should ask Oskar. > > The results seem pretty similar to the current method, at least in terms of > minimising the distance between neighbours (I haven't tested routing yet).
So a swap will only happen if it minimises stress for *both* nodes, rather than before it would happen if it minimised total stress for both nodes combined. Clearly this is a significant change to the transition probabilities... It might very well lead to false minima? One thing to try is to use the same random for each node's decision (since this is likely how it will be implemented in practice). In fact, lets be specific: If a swap will move B significantly closer to where it wants to be, and A slightly further away from where it wants to be, under the new system the probability of the swap depends only on A's reluctance. Under the old system, the swap would have gone ahead unconditionally. I'd expect this to be a lot slower, at best, and intractable at worst... but by all means simulate it. IMHO there may be an engineering solution to this (some form of self-regulation whereby locations are made public and swaps signed; entire bogus networks can then be handled by segregating them through network coloring), but we're some way away from being able to implement it. It's interesting to look at your logs for swap attempts. Set log level minor for freenet.node.Location and search for the regex "LocationManager.*, his:", and you will see that the swapping algorithm is definitely doing *something* : most nodes have mostly close peers and often a few further away. > > Cheers, > Michael -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 189 bytes Desc: not available URL: <https://emu.freenetproject.org/pipermail/devl/attachments/20080221/b0ced4ab/attachment.pgp>
