This dissertation describes Dynamic Multiversioning , a novel replication protocol, providing scalability, consistency and ease of reconfiguration for the back-end database in dynamic content servers. The key idea is the cooperation between a request scheduler, which orchestrates the replicated cluster, and a versioning-based replication technique integrated with the database fine-grained concurrency control engine. The scheduler distributes transactions on a set of lightweight database replicas and enforces serialization order by executing updates on a master node. The version-aware scheduling scheme guarantees strong 1-copy-serializability.Experiments with up to 9 database replicas show throughput scaling factors of 14.6, 17.6 and 6.5 respectively for the browsing, shopping and even for the write-heavy ordering workload of the industry-standard TPC-W benchmark. In addition, our technique guarantees that any crucial data can quickly and easily be recovered, which facilitates almost instantaneous reconfiguration, without loss of data, in the case of single-node failures of any node in the system.