Neil writes (with Ian's qualified agreement): > I think you put to much emphasis on the Inserts. > The success factor for Freenet's routing is the distributed cashing, the > fact that information migrates to where it is requested, this means that as > long as enough nodes look in the same place the information is most likely > going to be there(eventually). The fact that the insert and requests take > the same route helps the data to initially get found quicker.I think a lot > information will not stay where it gets inserted. > This is why I think the 'fireworks' aproach to updating is shaky, but also > why I think such a 'fuzzy' search aproach would succeed.
Even if we emphasize caching of retrievals, that won't make searches work. In the first place, some data won't be searched for very much, so it won't be widely cached. Freenet generally allocates space to data depending on how popular it is. Technical or obscure data will be on a relatively small number of nodes so it still won't be easy to find if the routing lacks precise targeting. In the second place, retrievals of data may be via different keywords. Going back to my example where the data is inserted with keywords A,B,C,D,E; one retrieval may be looking for B&D while another is looking for A&E, etc. These two searches won't go anywhere near each other. And in the third place, caching on retrievals will only spread the data if it is found. Someone still needs to find it that first time. If that step doesn't work, caching on retrievals won't help. I still don't see how the insert and search will find each other. Think of a network with 100,000 nodes, and the data you want has been inserted, but you are the first person fetching it. It lies on maybe 10 nodes in the network. There are 10,000 to 1 odds against any given node holding it. This means you need accurate routing if your search is going to find it in 10 hopes. With ordinary Freenet fetches, I can imagine that this will work, because we are zeroing in on the exact key that was inserted. But with searches, the location of the data depends on some information that the search-routing node doesn't have (i.e. keywords that were present on insert but aren't being used in the search). My concern is that the search routing algorithm doesn't have enough information to locate the original data, unlike with regular Freenet fetches. Hal _______________________________________________ Freenet-dev mailing list Freenet-dev at lists.sourceforge.net http://lists.sourceforge.net/mailman/listinfo/freenet-dev
