The document describes the evolution of OrientDB's database architecture from a master-slave model to a distributed, multi-master architecture. In the master-slave model, the single master node was a bottleneck for write operations. OrientDB's new architecture allows all nodes to accept writes, uses sharding to partition data across multiple nodes, and employs auto-discovery and replication techniques to improve scalability, availability, and failover. Hazelcast is used for distributed coordination and metadata storage.