Just curious, for node specialization, is the match width fixed? If the protocol was better documented I wouldn't have to ask.. :
I would think that it would be useful for a node to specify a mask width, i.e. a node with substantial resources could advertise a larger width. As it fills, it would throw out off center objects first (also biased with use of course) and reduce it's advertised width over time. Since much of freenet is probabilistic, it would be interesting to the width to describe the shape of a probability curve, which would be used for route selection and object storage. It might also be interesting to explore second degree specialization: Nodes which were inclined to maintain peering with similar specialized nodes, and who advertised the aggregated probability curve.. I am also curious, is the datastore dual queue for cached objects? It would make sense for objects which match the specialization to be kept in a separate replacement queue then objects which don't. The cache queue replacement policy would probably be more strongly based on hops to 'source' and a LFU moving average, while the longer term queue would be better weighted by specialization match and LRU. Just some random thoughts.. _______________________________________________ freenet-tech mailing list [EMAIL PROTECTED] http://lists.freenetproject.org/mailman/listinfo/tech
