This sounds like verifiable secret sharing with an honest majority. Here's a sampling of a few papers on related topics:
"Asynchronous Veriļ¬able Secret Sharing and Proactive Cryptosystems" http://eprint.iacr.org/2002/134.pdf "Distributed Private-Key Generators for Identity-Based Cryptography" http://www.cypherpunks.ca/~iang/pubs/DPKG-SCN10.pdf "Verifiable Secret Sharing and Multiparty Protocols with Honest Majority" http://cvs.cs.umd.edu/~gasarch/secretsharing/rabinVSS.pdf "Multiparty Computation with Faulty Majority" http://groups.csail.mit.edu/cis/pubs/shafi/1989-focs.pdf "Optimal Algorithms for Byzantine Agreement" http://dl.acm.org/citation.cfm?id=62225 On Thu, Jul 18, 2013 at 8:57 PM, Tony Arcieri <tony.arci...@gmail.com> wrote: > Has there been any work with combining Shamir-style secret sharing with > consensus protocols like Paxos and Raft (or leader election protocols like > Omega Meets Paxos)? > > The idea would be to have a network of n peers, who share a secret where t=2 > shares are required to reassemble the original secret. This secret is used > to sign new values when a group consensus is reached via a Paxos-like > protocol. > > In this scheme, a "proposer" would give its secret share, along with a > proposed new value, to "acceptor" nodes, who can reassemble the entire > secret. If they accept the new value, they can sign it with the secret, then > immediately erase it. If we use a deterministic signature algorithm like > Ed25519, every acceptor taking part in the consensus protocol can produce > the same signed version of the proposed new value. They can then continue > with the consensus protocol's accept phase. The result will be a quorum on a > signed value (or a consensus failure if quorum can't be reached, of course) > > Let's assume a malicious entity gains control of one and only one of the > nodes. They are now able to propose new values, so they can manipulate the > peer network by proposing malicious values which will get accepted by the > rest of the group. > > However, they do not *immediately* learn the private key. They would only > learn the private key if any other node were to propose a value which > contained their secret share. > > -- alternatively -- > > Secret sharing could be combined with a leader election protocol. In this > scheme, the leader and only the leader would learn the shared secret. All > proposed values would have to be approved and signed by the leader. > > I'm not sure I like this as much though. The leader is a single point of > failure, and an attacker could maliciously force a leader election through > e.g. DoS, having compromised only one other host directly. > > -- > Tony Arcieri > > _______________________________________________ > cryptography mailing list > cryptography@randombit.net > http://lists.randombit.net/mailman/listinfo/cryptography > _______________________________________________ cryptography mailing list cryptography@randombit.net http://lists.randombit.net/mailman/listinfo/cryptography