The subject of this research is ensuring data consistency and fault tolerance in distributed systems using consensus algorithms. The object of the study is the Paxos consensus algorithm, which is widely used to coordinate decisions between nodes in distributed environments. Despite its reliability, Paxos has several practical limitations that may affect system performance and availability under unstable network conditions. The paper analyzes several shortcomings of the classical Paxos algorithm, particularly the difficulty of gathering a quorum when some nodes fail or become temporarily unavailable, as well as performance degradation caused by the need to involve all acceptors in the consensus process. Special attention is paid to the influence of message loss and network instability on the efficiency of consensus algorithms in distributed systems. To address these challenges, the study proposes a modified version of the Paxos algorithm called Orchestra-Paxos. The proposed modification focuses on achieving partial consistency while improving system fault tolerance under conditions of failures and unstable network communication. In the proposed approach, the participation of more than half of the accepting nodes is not strictly required in each replication round. Consensus can be reached based on responses from an active subset of nodes, while other nodes are temporarily marked as unavailable. The research methodology is based on a comparative analysis of Orchestra-Paxos and Multi-Paxos using distributed system modeling and experimental performance evaluation. The experiments measure the average replication round time, the total completion time, and the number of rounds required to accept all proposals. The results show that Orchestra-Paxos significantly improves performance under high message loss conditions, reducing the required number of rounds and the overall completion time by an average factor of 12 compared to Multi-Paxos when the probability of message loss reaches 60%.
Diagilev et al. (Thu,) studied this question.