The Amaranth framework: Probabilistic, utility-based quality of service management for high-assurance computing

摘要

System resource management for high-assurance applications such as the command and control of a battle group is a complex problem. These applications often require guaranteed computing services that must satisfy both hard and soft deadlines. In addition, their resource demands can vary significantly over time with bursts of high activity amidst periods of inactivity. A traditional solution has been to dedicate resources to critical application tasks and to share resources among noncritical tasks. With the increasing complexity of high-assurance applications and the need to reduce system costs, dedicating resources is not a satisfactory solution. The Amaranth Project at Carnegie Mellon is researching and developing a framework for allocating shared resources to support multiple quality of service (QoS) dimensions and to provide probabilistic assurances of service. This paper is an overview of the Amaranth framework, the current results from applying the framework, and the future research directions for the Amaranth project.