Designing And Implementation Of Failure Detector In Asynchronous Distributed Systems

Fischer et al. had proved that there is no deterministic algorithm to solve consensus in asynchronous distributed systems, in which processes may crash. Hence Sam Toueg et al. had proposed unreliable failure detecttion technology. As a module, failure detector runs independently and outputs a list of suspected nodes. Other processes can access the list to judge whether the remote nodes are crrect or not.Using Object-Oriented Language, we designed and implemented a Failure Detection Service, namely FD={ d₁ , d₂,…,d_n}, which has modular architecture and is based on unreliable failure detector class◇P. The failure detection service module named d i running on the i th host serves all of local nodes (subscribers) with failure detection service, it follows that FD serve all nodes in the distributed system with detection service. FD establishes a mapping of a physical failure detector to several logical failure detectors, and each of ones is associated with a subscriber. As a result, FD is transparent to subscribers.FD consists of Object Interface module, Failure Detector module, Network Interface module and Object&Group Management module. Modules cooperate with each other and export logical failure detectors, which are associated with subscribers afterwards. The results of complete testing and performance analysis of FD are depicted in detail in the 3th chapter. Testing contents includes fds'correctnesss and robusticity itself (involving satisfying expectant objects or not, running correctly or not for ever in background, maintaining all kinds of data lists accurately or not, sending and receiving messages smoothly or not, etc.) and performance analysis of fds (involving resource consumption, time complexity, messages complexity, message bit complexity, the speed of convergence and the length of time to detect the crashed node, etc.). The data analysis in the experiment shows that this FD is practical and effective.