The Blockchain Of Oz : Specifying Blockchain Failures for Scalable Protocols Offering Unprecedented Safety and Decentralization

Abstract

Blockchains have starred an outstanding increase in interest from both business and research since Nakamoto’s 2008 Bitcoin. Unfortunately, many questions in terms of results that establish upper-bounds, and of proposals that approach these bounds. Furthermore, the sudden hype surrounding the blockchain world has led to several proposals that are either only partially public, informal, or not proven correct.

The main contribution of this dissertation is to build upon works that steer clear of blockchain puffery, following research methodology. The works of this dissertation converge towards a blockchain that for the first time formally proves and empirically shows deterministic guarantees in the presence of classical Byzantine adversaries, while at the same time pragmatically resolves unlucky cases in which the adversary corrupts an unprecedented percentage of the system. This blockchain is decentralized and scalable, and needs no strong assumptions like synchrony.

For this purpose, we build upon previous work and propose a novel attack of synchronous offchain protocols. We then introduce Platypus, an offchain protocol without synchrony. Secondly, we present Trap, a Byzantine fault-tolerant consensus protocol for blockchains that also tolerates up to less than half of the processes deviating. Thirdly, we present Basilic, a class of protocols that solves consensus both against a resilient-optimal Byzantine adversary and against an adversary controlling up to less than 2/3 of combined liveness and safety faults. Then, we use Basilic to present Zero-loss Blockchain (ZLB), a blockchain that tolerates less than 2/3 of safety faults of which less than 1/3 can be Byzantine. Finally, we present two random beacon protocols for committee sortition: Kleroterion and Kleroterion+ , that improve previous works in terms of communication complexity and in the number of faults tolerated, respectively.