As it is distributed, NEO is heavily based on network communications. Operations such as object load and store are done directly between client and storage node. The data flow is distributed among storage nodes to reduce the load and allow scalability.
Operations such as a transactional commit must be controlled by a central point, called the primary master node, as it guarantee unicity and atomicity. There is almost no cluster-wide operations, a transactions involves only a subset of all storage nodes, depending on the transaction content.