Toad wrote:
All in all, it could take a very long time for the estimators to start producing realistic estimates. If so, we should consider having a less dramatic bias towards untested nodes.
In which case they won't get routed to and we'll be stuck in false optima - but see my suggestion in the other mail.
No, I didn't say there would be no bias, just less of a bias.
And to which other mail are you referring?
This one, Ian.
Toad wrote:
Proposal:
We initialize the node estimators to something realistic, and not heavily biased. This may be from data from the seednodes or from the StoreData Source field, or we may make it up from averages etc.
We have a function E(node), which gives a rough estimate of the experience level of a node. When it is below some critical threshold, we randomly fork requests occasionally to that node, as that node won't be getting much other traffic (the established known good nodes will therefore get all requests - we don't sacrifice routing efficiency on a given request in order to test out a new node, we simply start another request occasionally). Could produce more traffic, but on the other hand the basic issue with traffic is routing effectiveness, and this might improve that and therefore thoroughly compensate for the extra traffic (which we would limit to some sane fraction)...
As no one has stepped up to shoot it down, I'll give it a tentative thumbs up.
-Martin
_______________________________________________ Devl mailing list [EMAIL PROTECTED] http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl
