While the potential applications of the consortium blockchain are becoming increasingly evident, the fault tolerance of systems in complex and changeable real environments has become an increasingly important factor. In recent years, many blockchain-oriented fault tolerance testing tools have been proposed. However, the selected fault sets cannot cover all kinds of problems that the consortium blockchain may encounter in real scenarios. Moreover, a rationality analysis of test results is often missing from these tools. In addition, it is also worth considering how to optimize system performance in the fault continuous stage. In this paper, we propose a general full-stack fault injection platform that can support the orderly injection of different kinds of Byzantine and non-Byzantine failures in a distributed scenario. Regarding the unclear underlying principles affecting system performance due to faults, we conduct an attribution analysis of various faults' influences on the consortium blockchain. Based on conclusions drawn form the attribution analysis, we design and implement the test-driven optimization strategy. The experimental results show that the optimization strategy can shorten the system average delay to less than one-third of that before optimization by reducing the throughput by about 15% in most fault scenarios.

Permissioned blockchain, as a multi-replica system, has its performance significantly affected by both the consensus protocol and the transaction execution protocol. Currently, there are many works optimizing the consensus or transaction execution of permissioned blockchain. However, existing works mainly focus on optimizing either consensus or transaction execution independently, lacking a holistic perspective. Based on this observation, we consider optimizing the permissioned blockchain from the holistic optimization perspective. Specifically, we heuristically design a transaction execution protocol TELL towards leaderless consensus to achieve collaborative optimization of consensus and transaction execution. Leaderless consensus is essentially a parallel running of multiple leader-based consensus instances, based on this characteristic, TELL pertinently designs intra-instance execution and inter-instances merging protocols. Additionally, we devise a novel State Hash Table (SHT) to record transactions' accessed states, so as to improve conflict serialization efficiency. Besides, we propose Dynamic Commitment Epoch (DCE) to adapt to instances' running status and decrease blocks' committing latency. Experimental results shows that compared with existing works, TELL further improves the performance of permissioned blockchain.