The Research On Consensus Protocols Based On A Failure Detector In Asynchronous System

The design and implement of fault-tolerant distributed applications is widely viewed as a complex endeavor. Requiring consistency of distributed applications make Consensus and Atomic Broadcast become key problems in research. In practice, Consensus can be used to elect a leader or agree on the value of a replicate data. Atomic Broadcast allows processes to agree on the set of messages they deliver and the order of broadcast messages. Considering different system models, although the asynchronous model is more attractive than synchronous model, FLP theory proves that Consensus and Atomic Broadcast cannot be solved deterministically in an asynchronous system in which processes may crash. Consensus is considered as a building block for distributed systems. Many researches have done to circumvent this impossibility. There are two main approaches, the first one uses random oracles and the second approach is based on equipping the purely asynchronous system with unreliable failure detection oracles. This paper mainly studies the second approach.Firstly it introduces definition and classification of the failure detectors and describes the method to implement a failure detector. Then analyzes the quality of service of the failure detector, this paper analyzes several kinds of typical consensus protocol based on the failure detector. The protocols including CT protocol, Paxos protocol, MR protocol and PR protocol. We use the NEKO framework to design the failure detector and the protocol. The Atomic Broadcast can be achieved by consensus, so the performance metric of consensus is the latency of Atomic Broadcast with the same throughput. We measure it with several different faultloads.In succession, this paper improves the consensus protocol based on the failure detectors through the property of spontaneous total order in local-area networks, the process is divided into two parts: the participant and the proposer, and the proposers broadcast messages in the first phase. Through experiments, the improved consensus protocol can reach a decision with one communication step if the proposers have the same value. By this way, it can enhance the efficiency of consensus protocol.At last, this paper introduces the adaptive failure detector and consensus protocol, brings forward the method of adjusting the membership in the group by the leader process Based on the round-trip delay matrix. Make those"fast"processes play a key role during the execution of the consensus protocols. This method can deal with the crash or running slowly of processes in the system.