PEM³ - the Policy Enhanced Memory Management Model

Over the past decade, clusters of workstations have become widely accepted as a cost efficient way of obtaining computational power. Moreover, clusters have increasingly been used to support multi-application environment, such as web servers and application servers, and to concurrently support a number of different services. In such an environment, resources become difficult to manage, e.g., it is difficult to adequately support the varying memory usage requirements of each application with a single strategy. In this paper we propose a policy-based model that provides applications with an interface to the underlying system in order to adapt the behavior of system services at runtime. The use of policies is illustrated by presenting the design of a memory management model for distributed shared memory systems, which allows different memory placement policies, while providing the ability to change consistency and coherency protocols at runtime.