An object-oriented framework for modular resource management

The authors present a flexible object-oriented framework for specifying modular resource management policies in concurrent systems. The framework generalizes the basic abstractions they originally developed for lottery scheduling. It is independent of the underlying proportional-share scheduler; a variety of probabilistic and deterministic algorithms can be used, including a min-funding revocation algorithm that they introduce for space-shared resources. The framework supports diverse resources and policies, including both proportional shares and guaranteed reservations. A repayment mechanism prevents allocation distortions caused by transfers of resource rights. Key framework concepts are analogous to features of object-oriented languages.