I think people get consensus finality mixed up with eventual consistency. (To be fair, they do sound like similar things.)<p>Nakamoto Consensus (used by Bitcoin and others) never reaches consensus finality, i.e. a point in time when you can be sure that consensus has been achieved. All you can do is estimate the probability that a block is in the final chain.<p>That said, Nakamoto Consensus has some really nice properties compared to classical consensus algorithms. For example, most "classical" BFT consensus algorithms have a worst-case message complexity of O(N^2) (or in some new highly-academic algorithms O(N polylog(N))). Nakamoto Consensus, in contrast, has O(N) worst-case message complexity. That's what enables Bitcoin (and the like) to scale so nicely to thousands of nodes.<p>Of course, if you're building a permissioned ledger with 25 nodes, that's irrelevant; the classical algorithms will work fine.<p>There's more discussion of this in the SCP paper: Luu, Loi, Viswesh Narayanan, Kunal Baweja, Chaodong Zheng, Seth Gilbert, and Prateek Saxena. "SCP: A Computationally-Scalable Byzantine Consensus Protocol For Blockchains."
This really highlights the differences between trying to understand and formalise the theory behind distributed systems and trying to practically build and use distributed systems.<p>In a practical sense, bitcoin pretty much has the property of strong consistency. That is a useful thing for people to know and understand. However, none of the theory that applies to strongly consistent systems necessarily <i>needs</i> to apply to bitcoin, because bitcoin is not theoretically perfectly strongly consistent. This is also important for people to understand, as otherwise they may incorrectly assume that certain results apply to bitcoin, e.g. the CAP theorem, leading to confusion.<p>So, I really do believe it is important to differentiate between talking about the properties of distributed systems in a theoretical sense and talking about them in a practical sense. The current formalisations of distributed systems have mainly been built to understand the theoretical properties, which while can be useful as a starting point to understand the practical properties, its always important to note that distributed systems may have good enough approximations to useful theoretical properties for a particular use case.
Like several others (who got downvoted, I don't know why) who have pointed out - this article provides no citations and confusing reasoning. The definition of "eventual consistency" is that all systems converge to the same state after enough time. Yet the author seems to be saying Bitcoin is strongly consistent because (if we ignore things in their variable state) it will converge to an agreed upon state.<p>Apparently it is up to you how you want to interpret the use of "strong" or "eventual" here.
Given that the blockchain can split at any time with a probability based on incentive that approaches infitesmal values as chain size goes to infinity, the blockchain is not even eventually consistent. It is never consistent, only probabilisticly consistent.
tldr: author claims if you only look at transactions with 6 or more confirmations, it's strong consistency (or rather, "your chances of observing an anomaly are exponentially small")<p>Anyone not using Satoshi's 6 confirmation rule is No true Scotsman.<p>No statistics are done on actual, real-life blockchain orphans to prove his claims.
How is omega not equivalent to the time interval in common EC strategies. Also, saying that "EC" doesn't guarantee serializability sounds strange. It does, over t. And yes, CAP is widely misunderstood, but I'm not sure this is the fault of the academics.<p>Sounds kind of like this was written to a certain audience that is in a bubble wrt the wider field.
> Because the probability drops exponentially with Ω, it's easy to pick an Ω such that the chance of observing a blockchain reorganization is less likely than having the processor miscompute values due to strikes by alpha particles. If that's not sufficient for pedants, one can pick a slightly higher Ω such that the chances of observing a reorganization are lower than the likelihood that all the oxygen molecules in the room, via random Brownian motion, will pile up into one corner of the room and leave one to suffocate in the other corner.<p>This guy obviously doesn't understand what he is talking about. I would more probably remine the whole bitcoin blockchain on my cpu in a second than this thing happens.