Experiences with Fault-Injection in a Byzantine Fault-Tolerant Protocol

摘要

The overall performance improvement in Byzantine fault-tolerant state machine replication algorithms has made them a viable option for critical high-performance systems. However, the construction of the proofs necessary to support these algorithms are complex and often make assumptions that may or may not be true in a particular implementation. Furthermore, the transition from theory to practice is difficult and can lead to the introduction of subtle bugs that may break the assumptions that support these algorithms. To address these issues we have developed Hermes, a fault-injector framework that provides an infrastructure for injecting faults in a Byzantine fault-tolerant state machine. Our main goal with Hermes is to help practitioners in the complex process of debugging their implementations of these algorithms, and at the same time increase the confidence of possible adopters, e.g., systems researchers, industry, by allowing them to test the implementations. In this paper, we discuss our experiences with Hermes to inject faults in BFT-SMaRt, a high-performance Byzantine fault-tolerant state machine replication library.