Impact of Communciation Networks on Fault-Tolerant Distributed Computing

摘要: When the desired reliability of a computing system exceeds that its individual hardware components need for fault-tolerant systems arise. While distributed have potential to achieve highly reliable computing, programming them is challenging task. Several paradigms been identified can simplify conceptual design systems. Properties profound implications on solvability and efficiency implementations these paradigms. In this thesis we study effect different communication models computing. As an instance fundamental operation examine protocols broadcast in Our main contribution characterization time complexity with respect models. A practical consequence our results development efficient variety common networks are shown support style communication. fact, by parameterizing minimum multicast size diameter networks, able characterize all known network architectures. Distributed where processors perceive same approximate makes much easier. Clock synchronization implement abstraction given only clocks bounded drift rates real time. We show how primitive which normally used also be synchronizing clocks. If occurs naturally sufficient frequency, clock achieved at no additional message cost. reveal hardware/software tradeoffs between performance, resiliency Thus, they offer many new alternatives previously not considered designing