Optimizing Split-Brain Arbitration Decisions in Dual Data Centers
摘要
The dual-active data center architecture is widely used in cloud computing to enhance system availability. However, this architecture also introduces risk of network partition failures that may cause system unavailability and data inconsistency. This paper proposes two optimized arbitration mechanisms to ensure data consistency and service recovery during split-brain failures. First, the Dynamic Arbitration Transfer Algorithm (DAT). It solves the single-point failure problem of the traditional single-arbitration node. This is achieved by using a three-zone arbitration cluster and implementing a distributed election mechanism. Second, the Pre-Arbitration Algorithm based on Consensus Mechanism (PACM). It uses the Raft consensus network to pre-synchronize the health scores and arbitration decisions, eliminating the dependence on third-party arbitration nodes and improving the arbitration efficiency. Experiments show that during split-brain failures, DAT and PACM outperform traditional arbitration algorithms, delivering significant improvements in data consistency, average system response time, and service recovery time. The research provides an efficient solution for data consistency and service recovery after split-brain failures in dual-active data center architectures. Moreover, these achievements offer valuable insights for achieving high availability and data consistency in distributed systems.