On 7/22/06, Julien Lociuro <[EMAIL PROTECTED]> wrote:
Ok, although I found an article describing atomic Put operation in a DHT. (It assumes nodes are not malicious, I think) Here it is : Etna : A Fault-Tolerant Algorithm fot Atomic Mutable DHT Data http://publications.csail.mit.edu/tmp/MIT-CSAIL-TR-2005-044.pdf
Thanks for the link. It looks like Etna uses quorums to provide atomicity, and that it uses the Paxos protocol to elect a leader for quorum decisions. Someone with a greater understanding of Paxos can correct me if wrong, but it does look like Paxos would be susceptible to malicious and misbehaving participants. From 'Paxos Made Simple': "We use the customary asynchronous, non-Byzantine model." In particular, the protocol states that an acceptor can only respond to an 'accept' request iff it has not promised not to. A malicious node that violates this part of the protocol breaks it. 'Acceptors' in Paxos that lie about values also break the protocol's guarantees. So, yeah, absolutely, if you can control all the nodes, or control all the software on all the nodes, it is a different ballgame. All the work I've seen (which is little) in this area uses quorum systems of one flavor or another. Alen _______________________________________________ p2p-hackers mailing list [email protected] http://lists.zooko.com/mailman/listinfo/p2p-hackers
