DAG-Sword: A Simulator of Large-Scale Network Topologies for DAG-Oriented Proof-of-Work Blockchains

TL;DR

This paper introduces DAG-Sword, a discrete-event simulator for large-scale DAG-based blockchain networks, demonstrating its ability to model realistic scenarios and evaluate protocol performance under malicious strategies.

Contribution

We present DAG-Sword, a novel simulator capable of modeling large-scale DAG-based blockchain networks with customizable topologies and realistic data, including malicious miner behaviors.

Findings

Malicious miners using different transaction strategies reduce protocol throughput.

Large-scale simulations with 7000 nodes extend prior small-scale results.

A new memory-efficient mempool structure improves simulation scalability.

Abstract

The blockchain brought interesting properties for many practical applications. However, some properties, such as the transaction processing throughput remained limited, especially in Proof-of-Work blockchains. Therefore, several promising directions, such as sharding designs and DAG-based protocols emerged. In this paper, we focus on DAG-based consensus protocols and present a discrete-event simulator for them. Our simulator can simulate realistic blockchain networks created from data of a Bitcoin network, while its network configuration and topology can be customized. The simulated network consists of honest and malicious miners. Malicious miners do not make any attack on consensus itself. Instead, they use a different transaction selection strategy than honest miners (who select transactions randomly) with the intention to earn unfairly more profits than honest miners at the cost of downgrading the protocol performance by duplicate transactions. As a consequence, this harms the performance of some DAG-based protocols (e.g., PHANTOM and GHOSTDAG) in terms of transaction processing throughput, which we demonstrate in our experiments and extend the results of the related work that contains a small-scale network of 10 nodes by the results obtained on a large-scale network with 7000 nodes. Next, we empirically compare different algorithms for the mempool structure, and we propose a composite mempool structure that is memory-efficient and thus convenient for simulations of resource-demanding large-scale networks.