On Friday 25 January 2008 01:24, Michael Rogers wrote: > Matthew Toseland wrote: > > I was trying to understand your proposal, I'm not saying I support it. The > > other option - for DNFs to be fatal - is much simpler. > > True, and per-node failure tables would probably solve the rabbit hole > problem. And I guess the requestHandler/Sender logic would be simpler if > we didn't have to think about moving on to other peers, and the timeouts > could probably be tighter.
Okay, so are we agreed on: - A pDrop of 5%, determined at each request/retry. - Some form of per-node failure tables (requires modifications to ULPRs). - Fatal DNFs. - On an RNF or an RO, we toss the coin again. As the optimal solution for the time being for replacing HTL and nearestLoc ? Main advantages: - nearestLoc-based attacks are eliminated. These are *nasty*! - Less effective destination samples: all samples being positive but low confidence actually minimises the information leaked. Beyond that obviously decreasing pDrop helps, but that's a security/load+time tradeoff. - Simple. - Predictable mean visited nodes of 1/pDrop. - Retried requests will go different ways. - We can fully explore small networks and small network pockets because we can retry on an RNF. Disadvantages: - Occasional timeouts due to requests going *a lot* of hops. - Need to retry all requests. Unclear: - 5% loss of the original "signal" per hop, in addition to the usual division of requests through routing and retrying. If this is the best solution, we can file a bug for it referencing this discussion and including its main conclusions, and probably implement it after 0.7.0 (soon). And move on to worrying about tunnels. > > 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/20080125/093bf382/attachment.pgp>
