Performance Modeling and Testing of DAG-based Distributed Ledger Systems

Performance is a major concern of the large-scale application of DLS (distributed ledger system). Compared with chain-based DLSs, DAG (directed acyclic graph) based DLSs are promising to enhance transaction parallel processing capabilities greatly and have gained increasing interest. However, due to the complex technology stack, current metrics, such as TPS (transactions per second) and latency, are insufficient for a deep understanding of DAG-based DLS performance. To address this problem, based on a comprehensive analysis of the transaction lifecycle process, we propose a state model and a set of performance indicators by identifying the key operations in the workflow. Then we develop an automated testing tool and conduct experiments on two representative open-source systems, IOTA and Conflux, considering their open-source nature, extensive documentation, and representativeness. The experiments profile the DAG-based DLS performance with respect to the system architecture, DAG topology, runtime behavior, and DAG processing mechanisms. The state model, indicators, and key experiment findings are valuable for future DLS design, deployment, and performance optimization.