The concurrency control problem in a federated database system (FDBS) is especially difficult due to the inherent heterogeneity and autonomy of participating local database systems. A number of FDBS concurrency control algorithms have been proposed. However, each algorithm has the drawbacks of low concurrency, global deadlocks or high system resource wastage. In this paper we propose a new protocol, called Dynamic Adjustment of Global Serialization Order (DAGSO), which provides high concurrency and is able to reduce system resource wastage due to early detection and abortion of eventually non-globally serializable transactions. The protocol is proved to be correct for achieving global serializability and free from global deadlocks. This paper also compares, using a detailed simulation model, the global transaction throughput of DAGSO with Top Down, Bottom Up, and Site Graph approaches proposed in the literature. The results show that DAGSO performs the best in most local system operating regions and range of global transaction behavior.