Server Power Management Through Dynamic Job Classification

 Technology Overview: 

With the growing concern about energy expenditures in server installations, mechanisms are needed for improving the energy-efficiency of server installations. A technique is proposed for learning the energy-performance characteristics of jobs or phases of jobs as they execute. This information is used to schedule new jobs as well as repetitive jobs with the server characteristics set dynamically to meet the performance demands using the most energy-efficient configuration for realizing the performance goal. The proposed technique also amortizes the energy and performance overhead for dynamically changing the energy-performance setting of the server.

 Advantages: 

This technique has the potential for improving the energy-efficiency of server systems through relatively modest changes in the operating system kernel. No reliance is made on unique hardware or dedicated hardware support.