Consensus-less Security

Abstract

Distributed ledger technology was expected to spark a technical revolution similar to the internet revolution. After the release of Bitcoin in 2008, many developments have significantly increased the performance of distributed ledger technology. Nevertheless, the first truly scalable ledger has yet to be deployed. All of them have issues with scaling in either the throughput, the number of nodes which can validate transaction or both. The concept behind a distributed ledger is that the integrity of the ledger is a shared responsibility. However, as soon as new technology emerges, also misuse surfaces, especially if there are financial gains involved. The general solution, to prevent such abuse, in distributed ledger technology is through the use of global consensus. If the majority of a network is honest, and we require a majority vote on the validity of a transaction, no malicious transactions will succeed. A downside of requiring a majority vote is that every node eligible to vote must contain full knowledge on all previous transactions. This work argues that the requirement of global consensus is a major limiting factor when it comes to the scalability of current ledgers. The goal of this work is to design a scalable distributed ledger whose security does not rely on global consensus. It proposes a novel algorithm that guarantees security, even under adversarial attack, by up to a third of the network exhibiting byzantine behavior. It does so using Trustchain, a pair-wise ledger designed by the Delft University of Technology, and `Fair Witness Selection Protocol', a newly designed publicly verifiable witness selection algorithm with an indicated message and communication complexity of $O(log^\star(n))$. A mathematical lower-bound is given on the security level of the algorithm, and the security is reduced to the security of the underlying hash function. Several experiments were executed on the DAS-5 supercomputer to confirm the scalability of this work. These experiments show that the throughput of the network scales linearly, and has been tested up to 2500 nodes (simultaneously acting as validators and clients). To the best of the author's knowledge, it is the only ledger that has no theoretical limits on the number of clients, number of validators, or throughput. A peak-throughput of 7025 tx/s has been observed at a network size of 280 nodes. Furthermore, the total transaction time remained roughly constant at about 15 milliseconds regardless of the network size.