A machine's physical memory is a resource for which many subsystems, including databases, persistent object storage managers, and multimedia file systems, are potential competitors. As memory size grows, optimizing the diverse uses of physical memory will require diverse management policies. These policies must, however, coexist, and the claims on memory for different uses must be balanced. This research investigates a system architecture for balancing the requirements of memory competitors, while allowing a greater variety of management policies to be implemented. The system architecture permits pageout daemons situated outside the kernel to control pools of physical memory. This allows subsystems such as databases much greater control over memory management policy. A balance manager within the kernel controls the size of these pools. A simulator is used to explore the interactions of management policies, while a prototype system, supported by an extended version of Mach 3.0, allows the experimental validation of the simulation results. Successful completion of this project will provide a framework for managing large physical memories in future machines. Using the balancing strategies identified in this research, subsystems for memory- intensive application will be more readily developed and installed on new machines without ad hoc performance tuning.
