Due to the recent improvements in the price/performance characteristics of workstations and the networking capabilities, the client-server system architecture has become a target for database systems. A client-server database management system (DBMS) provides the management of a database that resides on a client-server system. Data-access requests of the clients are handled by the database servers. The whole database is stored on the disks that can be accessed by only the servers, and in order to reduce disk accesses, copies of database items can be cached in the global memory which comprises the memories of all the computers connected to the system. Designing efficient global memory management algorithms helps the transactions experience less disk input/output (I/O) during their execution. In this paper, we propose a global memory management algorithm for client-server DBMSs which aims to reduce disk I/O by increasing the portion of the database available in global memory. Performance of the algorithm is examined by a comparison with some previously proposed algorithms, using a simulation model designed for studying various performance issues in client-server DBMSs. Conditions for which the new algorithm provides significant improvements in the overall throughput of the system are identified.