We introduce a Resource Access Management (RAM) framework for resource utilities that facilitates Class of Service (CoS) based automated resource management. The framework may be used to offer resources on demand to enterprise applications that have time varying resource needs. The classes of service include guaranteed, predictable best effort, and best effort. The analytical apparatus we exploit requires the notion of application demand profiles that specify each application's resource requirements. These profiles may be statistical in nature. Consequently a policing mechanism is introduced to constrain each application's resource usage within its profile. A case study that exploits data from 48 data center servers, is used to demonstrate the framework. We show that our techniques are effective in: exploiting statistical multiplexing while providing service level assurances, limiting application demands in the presence of hostile application behaviour, and providing for differentiated service levels as planned.
展开▼
