• foota a day ago |
    Ok, I'm a bit surprised to see no comments here, I guess everyone is waiting to read the paper till after the work day? :)

    I thought it was interesting, iiuc the proposal is that you can design a protocol like Paxos that allows you to guarantee that some replica will win in the future, I'm not sure how useful this is in practice though, since you won't know that your (you being the client) writes are winning until after the fact, and no replica could give you the true state until convergence. I find the claim that this represents partial progress slightly misleading, but interesting nonetheless.

  • sargun a day ago |
    Wasn't there a paper from Heidi Howard that stated you it's not actually f failures in 2f+1 nodes, but you can actually make the number of nodes that can fail a lot higher, if you have some nodes that have to take part in the quorum (or the prior proposal)? This feels roughly similar?
    • foota a day ago |
      I think that's a different thing... I believe you're talking about flexible paxos (https://arxiv.org/abs/1608.06696).

      Flexible paxos doesn't change the underlying guarantees at all, whereas this does.