Please note that this seminar has been cancelled
Yuhao (Eric) Dong,
PhD
candidate
David
R.
Cheriton
School
of
Computer
Science
Consensus algorithms used to secure public blockchains differ significantly in their design constraints from those driving traditional fault-tolerant distributed systems. This is because blockchains must rigorously model trust within a game-theoretical model, rather than simply making assumptions about fault tolerance. To truly achieve incentive-compatible security, we cannot rely on the typical approach of considering ideal honest behavior and then positing an adversary with certain powers. Unfortunately, game-theoretical analysis of multi-party coordination problems, of which blockchain consensus is an instance, tends to be pernicuously difficult, leading to most consensus algorithms in use lacking formal analysis.
Synkletos is a blockchain consensus algorithm designed through a novel approach that drastically simplifies incentive analysis. Instead of modeling the ideal blockchain as decentralized parties participating in a coordination game to produce a certain optimal behavior, we start by proposing an ideal monopoly blockchain, where blockchain rules are such that even a selfish monopoly will behave in a “faulty” way. We then design an simple incentive structure and consensus algorithm based on a variation of proof of stake that incentivizes any number of uncoordinated or coordinated parties to simulate such a monopoly as a whole. We argue that such a mechanism, though it is slightly less economically efficient compared to systems such as Nakamoto consensus that rely on noncoordination assumptions, is much easier to analyze and far more robust to game-theoretical attacks.