On 2012-07-20 3:43 AM, Tony Arcieri wrote:
To confuse the algorithm you would either need an extremely small pool of peers (e.g. you're still being bootstrapped into the network and you don't know enough to tell good peers from bad), or a large number of lying-but-otherwise-good-faith participants to convince a given peer to interact with a Sybil.
Someone who gives every man his due, which is to say a peer that supplies bandwidth and storage in return for bandwidth and storage, is likely to tell the truth about other peers, though by no means guaranteed.
Determining truth by consensus, and filtering out those that deviate from consensus, is apt to go into a positive feedback loop that winds up dominated by lying sybils.
This problem can be mitigated by tracing every purported fact to the original alleged observer, and ensuring that you are connected to the original observer by a chain of known good people: You have reason to trust Bob because Bob gave you your due, and reason to trust Carol, because Bob says that Carol gave him her due, and reason to trust Dave, because Carol says Dave gave her his due. Of course as the chain gets longer, one's confidence in the chain exponentially approaches zero, but as the chain has more links in parallel, one's confidence in the chain exponentially approaches unity.
You have reason to trust Bob and Carol because they both gave you your due, and reason to trust Dave because both Bob and Carol say Dave gave them their due. This is almost the same as the "determine truth by consensus" rule, but unlike the consensus rule, free from positive feedback loops, provided one uses Bayesian probability throughout.
This difference between the quite disastrous consensus rule, and this rule, is the correct use of Bayesian probability.
_______________________________________________ p2p-hackers mailing list [email protected] http://lists.zooko.com/mailman/listinfo/p2p-hackers
