> FROM: Scott G. Miller > DATE: 04/19/2000 14:46:45 > SUBJECT: RE: [Freenet-dev] Proposal for the Near Future > (Searching, CHKs > > I`m not sure if I`m following exactly, but it sounds like you > want to > start storing data on nodes by keyword.
That is the gist. > Its a *really* bad idea to > cluster data with the same keyword on the same node. First, an > attacker > need only bring down a handfull of nodes to destroy most of the > data > relating to that key. You are making the extreme assumption in this case that the routing mechanism clusters the vast majority of of the keyword hashes into a single node or network. This may be the case now but don't not necessarily need to be the case in the future. This clustering all depends on the routing comparison method (which IPs have affinity for which hashes) and this is definitely maleable. I have mentioned some of the ways that this can be changed in another post. The more I think about it though the more that it will not be as big an issue as people make it out to be. Data gets cached all the way into the focus of the routing. When people request this data, the requests get satified *long* before the request reaches the focus node. In fact, I'll bet that the data lasts a lot longer on nodes that are not as good a routing match as *the best* routing match at the focus of the request. The reason for this is that the requests are going to be satisfied (therefore voted for) way more times on nodes on the route to the focus then by the focus node itself. Therefore, ther more popular a particular keyword becomes, the more distributed the keyword hash +data becomes, the larger the ring of nodes enroute to the focus becomes that satisfy the request before it reaches the focus. This whole freenet caching mechanism reveals more and more elegance as time goes on. The routing could use a little damping in it though to avoid the whole "find focus too soon and orbit" scenario that I have described in my other posts today. > > Secondly, if that node goes down you loose a lot of data, Not so due to aformentioned scattering of data. > and finally it > doesnt distribute the load for that type of request around the > network. I think you may be refering to the "general requests" for metadata that I described. True, these will not distribute data. However, I also mentioned that when one makes a final "specific request" for a particular piece of data, the client should also make a few specific requests for the metadata that the user deemed relevent to their decision. In that way, the best metadata survives. Otherwise, all the pure metadata documents would die in this scheme. > > I apologize if this wasnt what you meant. Discussion and critical analysis is what this is all about, no? I would much rather you raise doubts and questions and shoot holes in my zany ideas than ignoring them out-right. Thanks for reading my long-winded postings. (and appearently long lined in some email clients ... I didn't realize that my line wrapping was disfunctional since I normally read the postings on the web) Mike _______________________________________________ Freenet-dev mailing list Freenet-dev at lists.sourceforge.net http://lists.sourceforge.net/mailman/listinfo/freenet-dev
