tailieunhanh - A Comparison of High-Level Full-System Power Models

Pressing demand. Past market-based systems never saw real field testing, and contention was often artifi- cially generated. Today, a deployed market system could have immediate usage and solve real resource conflicts. Real usage data will help researchers calibrate and eval- uate their market-based resource schedulers. Previous mechanism designs were not able to take advantage of user feedback to drive the mechanism design process. Improved operating system infrastructure. Past sys- tems had to deal with limitations in infrastructure, such as a lack of user authentication or kernel-supported re- source isolation. Today, systems research has produced tools like BSD Jails, Xen, and Linux CKRM [21, 22], which are already in use to provide resource isolation, can. | A Comparison of High-Level Full-System Power Models Suzanne Rivoire Parthasarathy Ranganathan Christos Kozyrakis Sonoma State University Hewlett-Packard Labs Stanford University Abstract Dynamic power management in enterprise environments requires an understanding of the relationship between resource utilization and system-level power consumption. Power models based on resource utilization have been proposed in the context of enabling specific energy-efficiency optimizations on specific machines but the accuracy and portability of different approaches to modeling have not been systematically compared. In this work we use a common infrastructure to fit a family of high-level full-system power models and we compare these models over a wide variation of workloads and machines from a laptop to a server. This analysis shows that a model based on OS utilization metrics and CPU performance counters is generally most accurate across the machines and workloads tested. It is particularly useful for machines whose dynamic power consumption is not dominated by the CPU as well as machines with aggressively power-managed CPUs two classes of systems that are increasingly prevalent. 1 Introduction In order to maximize energy efficiency whether in a single system or over an ensemble of systems users and data center operators need to understand the relationship between resource usage and system-level power consumption. This understanding enables such optimizations as consolidating workloads on as few machines as possible and turning others off 6 16 since current hardware is highly inefficient at low utilization 1 . It also includes dynamically adjusting power budgets within an enclosure or rack to enable a data center s power provisioning infrastructure to be designed less conservatively 5 18 . While several types of full-system power models have been proposed often in the context of enabling a particular optimization they have not been systematically compared over a variety of .