The Implementation On Failure Detector In Byzantine Fault Tolerance Replication System

With the growing prevalence and development of Internet technology, the computer plays an important role in our daily lives. There is increasing need to build systems that provide high reliability to ensure correctness, and high availability for uninterrupted service. Recent work on Byzantine fault tolerant (BFT) state machine systems has demonstrate that generalized state machine replication can be used to improve reliability and availability. There is a large amount of research about BFT state machine. Some research has focused on high confidentiality against malicious attack to steal data. Some research has focused on high throughput to keep pace with high system load. And some research has focused on high availability to improve the performance. The paper explores how to improve the availability of BFT system by using failure detector. The main research results are listed below:Firstly, the paper argues for improving on adaptable failure detector algorithm to ensure the quality of service of failure detector not only for message delay but also for message lost. The algorithm defines the minimum value of safety margin of timeout delay and updates the freshness point when the state of failure detector is being changed from S-transition to T-transition. These methods reduce the mistake number and improve on mistake recurrence time. The ameliorative algorithm proposes that different implementation must use different number of lost messages for the different requirement of the quality of service of failure detector.Secondly, the paper describes using failure detector in BFT state machine replication architectures to shift view faster by failure detector. When continuous primary replicas are fault, BFT system can't reply requests in a long time. The work of failure detector is detecting failure timely. When primary is crash-stop failure and can be detected by failure detector, the result of failure detector can help BFT to shorten waiting time about view change. When continuous primary replicas are crash-stop failure, failure detector can shorten more waiting time.Thirdly, the paper implements adaptable failure detector in BFT system. The interface of failure detector and BFT library can prevent Byzantine failure from forging heartbeat messages. The load of message is trivial by reducing a great deal of detected replicas. When continuous primary replicas are crash-stop failures, the results of evaluation indicate that failure detector provides orders of magnitude improvements in availability over BFT library by reducing the cost of view change.