Consensus Algorithms for Distributed Systems: Managing the Consistency of Critical Data Files

This thesis investigates consensus algorithms for distributed systems, with a focus on their application in managing the consistency of critical data files. The work originates from a real-world project aimed at developing a distributed storage service based on a star topology network, with a central client replicating a file across multiple independent servers. After analysing the theoretical foundations of distributed systems and consensus algorithms, including the CAP theorem and the Fischer–Lynch–Paterson impossibility result, the study identifies the key principles that guide the design of practical consensus protocols. Among the existing approaches discussed in the literature, the Raft algorithm was selected as the most suitable reference model. Based on these principles, a custom consensus algorithm was designed for the system under study, adapting Raft to the architectural constraints imposed by the star topology. The algorithm was validated through simulations under various failure scenarios. As a further contribution, a reputation-based extension was introduced to improve the quality of decisions when consensus could not be reached, showing how reputation systems can enhance the reliability of distributed consensus.